feat:first commit

LICENSE.txt

@@ -0,0 +1,674 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.

README.md

@@ -0,0 +1,26 @@
+# AngioPy Segmentation
+
+## AngioPy paper
+Please see [here](https://doi.org/10.1016/j.ijcard.2024.132598) for our paper in the International Journal of Cardiology
+
+Please cite it!
+> Mahendiran, T., Thanou, D., Senouf, O., Jamaa, Y., Fournier, S., De Bruyne, B., ... & Andò, E. (2025). AngioPy Segmentation: An open-source, user-guided deep learning tool for coronary artery segmentation. International journal of cardiology, 418, 132598.
+
+## AngioPy in the news
+AngioPy segmentation is being used in [this RTS reportage on AI in Cardiology](https://www.rts.ch/play/tv/19h30/video/lia-fait-irruption-en-cardiologie-et-redefinit-le-role-des-medecins?urn=urn:rts:video:15479233) (in French)
+
+## Online Example
+Please visit https://imaging.epfl.ch/angiopy-segmentation/ for a live demo of this code on some example DICOM images
+
+![](illustration.mp4)
+
+## Description
+This software allows single arteries to be segmented given a few clicks on a single time frame with a PyTorch 2 Deep Learning model.
+
+## Installing and running
+ - Install dependencies: ` pip install -r requirements.txt`
+ - Launch Streamlit Web Interface: `streamlit run angioPySegmentation.py --server.fileWatcherType none`
+
+ ...a website should pop up in your browser!
+
+ You need to create a /Dicom folder and put some angiography DICOMs in there

angioPyFunctions.py

@@ -0,0 +1,265 @@
+import numpy
+import scipy.interpolate
+import skimage.filters
+import skimage.morphology
+import scipy.ndimage
+import scipy.optimize
+import predict
+from PIL import Image
+from fil_finder import FilFinder2D
+import astropy.units as u
+from tqdm import tqdm
+import pooch
+import utils.dataset
+import cv2
+
+
+colourTableHex = {
+                'LAD':       "#f03b20",
+                'D':         "#fd8d3c",
+                'CX':        "#31a354",
+                'OM':        "#74c476",
+                'RCA':       "#08519c",
+                'AM':        "#3182bd",
+                'LM':        "#984ea3",
+                }
+
+colourTableList = {}
+
+for item in colourTableHex.keys():
+    ### WARNING HACK: The colours go in backwards here for some reason perhaps related to RGBA?
+    colourTableList[item] = [int(colourTableHex[item][5:7], 16),
+                             int(colourTableHex[item][3:5], 16),
+                             int(colourTableHex[item][1:3], 16)]
+
+
+def skeletonise(maskArray):
+    # if len(maskArray.shape) == 3:
+    maskArray = cv2.cvtColor(maskArray, cv2.COLOR_BGR2GRAY)
+
+    skeleton = skimage.morphology.skeletonize(maskArray.astype('bool'))
+
+    # Process the skeleton and find the longest path
+    fil = FilFinder2D(skeleton.astype('uint8'),
+                    distance=250 * u.pc, mask=skeleton, beamwidth=10.0*u.pix)
+    fil.preprocess_image(flatten_percent=85)
+    fil.create_mask(border_masking=True, verbose=False,
+                    use_existing_mask=True)
+    fil.medskel(verbose=False)
+    fil.analyze_skeletons(branch_thresh=400 * u.pix,
+                        skel_thresh=10 * u.pix, prune_criteria='length')
+
+    # add image arrays dictionary
+    # tifffile.imwrite(os.path.join(arteryFolder, "skel.tif"), fil.skeleton.astype('<u1')*255)
+
+    skel = fil.skeleton.astype('<u1')*255
+
+    return skel
+
+
+def skelEndpoints(skel):
+    #skel[skel!=0] = 1
+    skel = numpy.uint8(skel>0)
+
+    # Apply the convolution.
+    kernel = numpy.uint8([[1,  1, 1],
+    [1, 10, 1],
+    [1,  1, 1]])
+    src_depth = -1
+    filtered = cv2.filter2D(skel,src_depth,kernel)
+
+    # Look through to find the value of 11.
+    # This returns a mask of the endpoints, but if you
+    # just want the coordinates, you could simply
+    # return np.where(filtered==11)
+    out = numpy.zeros_like(skel)
+    out[numpy.where(filtered==11)] = 1
+    endCoords = numpy.where(filtered==11)
+    endCoords = list(zip(*endCoords))
+    startPoint = endCoords[0]
+    endPoint = endCoords[1]
+
+    # print(f"Skel starts at {startPoint} and finishes at {endPoint}")
+
+    return startPoint, endPoint
+
+
+def skelPointsInOrder(skel, startPoint=None):
+    """
+    put in a skel image, get the y, x points out in order
+    """
+
+    # Lazy!!
+    if startPoint is None:
+        startPoint, _ = skelEndpoints(skel)
+
+    # get the coordinates of all points in the skeleton
+    skelXY = numpy.array(numpy.where(skel))
+    skelPoints = list(zip(skelXY[0], skelXY[1]))
+    skelLength = len(skelPoints)
+
+    # Loop through the skeleton starting with startPoint, deleting the starting point from the skelPoints list, and finding the closest pixel. This is appended to orderedPoints. startPoint now becomes the last point to be appended.
+    startPointCopy = startPoint # copied as we are going to loop and overwrite, but want to also keep the original startPoint
+    orderedPoints = []
+
+    while len(skelPoints) > 1:
+
+        skelPoints.remove(startPointCopy)
+
+        # Calculate the point that is closest to the start point
+        diffs = numpy.abs(numpy.array(skelPoints)-numpy.array(startPointCopy))
+        dists = numpy.sum(diffs,axis=1) #l1-distance
+        closest_point_index = numpy.argmin(dists)
+        closestPoint = skelPoints[closest_point_index]
+        orderedPoints.append(closestPoint)
+
+        startPointCopy = closestPoint
+
+    orderedPoints = numpy.array(orderedPoints)
+
+    # YX points
+    return orderedPoints
+
+
+def skelSplinerWithThickness(skel, EDT, smoothing=50, order=3, decimation=2):
+    # NOTE: the coordinate seem to come out with y first, then x
+    startPoint, endPoint = skelEndpoints(skel)
+
+    # Impose an order to points
+    orderedPoints = skelPointsInOrder(skel, startPoint)
+
+    # unzip ordered points to extract x and y arrays
+    x = orderedPoints[:, 1].ravel()
+    y = orderedPoints[:, 0].ravel()
+
+    x = x[::decimation]
+    y = y[::decimation]
+
+    #NOTE: Should the EDT be median filtered? I wonder in fact if doing so will reduce the accuracy of the model.
+    # EDT = skimage.filters.median(EDT)
+
+    t = EDT[y, x]
+
+    x = x[0:-1]
+    y = y[0:-1]
+    t = t[0:-1]
+
+    print(x.shape, y.shape, t.shape)
+
+    tcko, uo = scipy.interpolate.splprep(
+        [y, x, t], s=smoothing, k=order, per=False)
+
+    return tcko
+
+
+def arterySegmentation(inputImage, groundTruthPoints, segmentationModelWeights=None):
+    """
+    Segment a single greyscale artery with a UNet model.
+
+    Parameters
+    ----------
+        inputImage: 2D numpy array
+            Ideally this input is normalised 0-255 and 512x512
+            If a different size it is rescaled along with groundTruthPoints
+
+        groundTruthPoints: Nx2 numpy array
+            Y and X positions of annotated points along the artery,
+            Ordering is not important except that start and end points should be top and bottom of the array
+
+        segmentationModelWeights: segmentation model weights (pth), optional
+            Segmentation model weights to use.
+            If not set the default ones from this paper: https://doi.org/10.1016/j.ijcard.2024.132598
+
+    Returns
+    -------
+        mask : 512x512 numpy array (int64)
+            Mask selecting the selected artery, 0 = background and 1 = artery
+    """
+    if segmentationModelWeights is None:
+        segmentationModelWeights = pooch.retrieve(
+            url="doi:10.5281/zenodo.13848135/modelWeights-InternalData-inceptionresnetv2-fold2-e40-b10-a4.pth",
+            known_hash="md5:bf893ef57adaf39cfee33b25c7c1d87b",
+        )
+
+    if inputImage.shape[0] != 512 and inputImage.shape[1] != 512:
+        ratioYX = numpy.array([512./inputImage.shape[0], 512./inputImage.shape[1]])
+        print(f"arterySegmentation(): Rescaling image to 512x512 by {ratioYX=}, and also applying this to input points")
+        inputImage = scipy.ndimage.zoom(inputImage, ratioYX)
+        points = groundTruthPoints.copy() * ratioYX
+        print(inputImage.shape)
+    else:
+        points = groundTruthPoints
+
+    imageSize = inputImage.shape
+
+    n_classes = 2 # binary output
+
+    net = predict.smp.Unet(
+        encoder_name='inceptionresnetv2',
+        encoder_weights="imagenet",
+        in_channels=3,
+        classes=n_classes
+    )
+
+    net = predict.nn.DataParallel(net)
+
+    device = predict.torch.device('cuda' if predict.torch.cuda.is_available() else 'cpu')
+    net.to(device=device)
+
+    net.load_state_dict(
+        predict.torch.load(
+            segmentationModelWeights,
+            map_location=device
+        )
+    )
+
+    orig_image = Image.fromarray(inputImage)
+
+    image = predict.Image.new('RGB', imageSize, (0, 0, 0))
+    image.paste(orig_image, (0, 0))
+
+    imageArray = numpy.array(image).astype('uint8')
+
+    # Clear last channels
+    imageArray[:, :, -1] = 0
+    imageArray[:, :, -2] = 0
+
+    ## Get endpoints of skeleton
+    startPoint = points[0]
+    endPoint = points[-1]
+
+    # End points on Channel 1
+    for y, x in [startPoint, endPoint]:
+        y = int(numpy.round(y))
+        x = int(numpy.round(x))
+        imageArray[y-2:y+2, x-2:x+2, 1] = 255
+
+    # All other points on Channel 2
+    for y, x in points[1:-1]:
+        y = int(numpy.round(y))
+        x = int(numpy.round(x))
+        imageArray[y-2:y+ 2, x-2:x+2, 2] = 255
+
+    image = Image.fromarray(imageArray.astype(numpy.uint8))
+
+    mask = predict.predict_img(
+        net=net,
+        dataset_class=utils.dataset.CoronaryDataset,
+        full_img=image,
+        scale_factor=1,
+        device=device
+    )
+
+    return mask
+
+
+
+def maskOutliner(labelledArtery, outlineThickness=3):
+
+    # Compute the boundary of the mask
+    contours, _ = cv2.findContours(labelledArtery, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
+    tmp = numpy.zeros_like(labelledArtery)
+    boundary = cv2.drawContours(tmp, contours, -1, (255,255,255), outlineThickness)
+    boundary = boundary > 0
+
+    return boundary

angioPySegmentation.py

@@ -0,0 +1,335 @@
+import os
+import os.path
+import matplotlib.pyplot as plt
+import numpy
+import pandas as pd
+import streamlit as st
+import SimpleITK as sitk
+import pydicom
+import glob
+import mpld3
+import streamlit.components.v1 as components
+import plotly.express as px
+import plotly.graph_objects as go
+import tifffile
+from streamlit_plotly_events import plotly_events
+from streamlit_drawable_canvas import st_canvas
+from PIL import Image
+# from streamlit_image_coordinates import streamlit_image_coordinates
+import predict
+import angioPyFunctions
+import scipy
+import cv2
+
+import ssl
+
+ssl._create_default_https_context = ssl._create_unverified_context
+
+st.set_page_config(page_title="AngioPy Segmentation", layout="wide")
+
+if 'stage' not in st.session_state:
+    st.session_state.stage = 0
+
+
+
+# Make output folder
+# os.makedirs(name=outputPath, exist_ok=True)
+
+# arteryDictionary = {
+#     'LAD':       {'colour': "#f03b20"},
+#     'CX':        {'colour': "#31a354"},
+#     'OM':    {'colour' : "#74c476"},
+#     'RCA':       {'colour': "#08519c"},
+#     'AM':   {'colour' : "#3182bd"},
+#     'LM':        {'colour' : "#984ea3"},
+# }
+
+# def file_selector(folder_path='.'):
+#     fileNames = [file for file in glob.glob(f"{folder_path}/*")]
+#     selectedDicom = st.sidebar.selectbox('Select a DICOM file:', fileNames)
+#     if selectedDicom is None:
+#         return None
+
+#     return selectedDicom
+
+@st.cache_data
+def selectSlice(slice_ix, pixelArray, fileName):
+
+    # Save the selected frame 
+    tifffile.imwrite(f"{outputPath}/{fileName}", pixelArray[slice_ix, :, :])
+
+    # Set the button as clicked
+    st.session_state.btnSelectSlice = True
+
+
+DicomFolder = "Dicoms/"
+# exampleDicoms = {
+#     'RCA2' : 'Dicoms/RCA1',
+#     'RCA1' : 'Dicoms/RCA4',
+#     # 'RCA2' : 'Dicoms/RCA2',
+#     # 'RCA3' : 'Dicoms/RCA3',
+#     # 'LCA1' : 'Dicoms/LCA1',
+#     # 'LCA2' : 'Dicoms/LCA2',
+# 
+# }
+exampleDicoms = {}
+files = sorted(glob.glob(DicomFolder+"/*"))
+for file in files:
+    exampleDicoms[os.path.basename(file)] = file
+
+
+# Main text
+st.markdown("<h1 style='text-align: center;'>AngioPy Segmentation</h1>", unsafe_allow_html=True)
+st.markdown("<h5 style='text-align: center;'> Welcome to <b>AngioPy Segmentation</b>, an AI-driven, coronary angiography segmentation tool.</h1>", unsafe_allow_html=True)
+st.markdown("")
+
+# Build the sidebar
+# Select DICOM file: here eventually we will use the file_uploader widget, but for the demo this is deactivate. Instead we will have a choice of 3 anonymised DICOMs to pick from
+# selectedDicom = st.sidebar.file_uploader("Upload DICOM file:",type=["dcm"], accept_multiple_files=False)
+
+# def changeSessionState():
+
+#     # value += 1
+
+#     print("CHANGED!")
+
+
+DropDownDicom = st.sidebar.selectbox("Select example DICOM file:",
+                        options = list(exampleDicoms.keys()),
+                        # on_change=changeSessionState(st.session_state.key),
+                        key="dicomDropDown"
+                    )
+
+selectedDicom = exampleDicoms[DropDownDicom]
+
+stepOne = st.sidebar.expander("STEP ONE", True)
+stepTwo = st.sidebar.expander("STEP TWO", True)
+
+# Create tabs 
+tab1, tab2 = st.tabs(["Segmentation", "Analysis"])
+
+# Increase tab font size
+css = '''
+<style>
+    .stTabs [data-baseweb="tab-list"] button [data-testid="stMarkdownContainer"] p {
+    font-size:16px;
+    }
+</style>
+'''
+
+st.markdown(css, unsafe_allow_html=True)
+
+# while True:
+# Once a file is uploaded, the following annotation sequence is initiated 
+if selectedDicom is not None:
+        try:
+            print(f"Trying to load {selectedDicom}")
+            dcm = pydicom.dcmread(selectedDicom, force=True)
+
+            # handAngle = dcm.PositionerPrimaryAngle
+            # headAngle = dcm.PositionerSecondaryAngle
+            # dcmLabel = f"{'LAO' if handAngle > 0 else 'RAO'} {numpy.abs(handAngle):04.1f}° {'CRA' if headAngle > 0 else 'CAU'} {numpy.abs(headAngle):04.1f}°"
+
+            pixelArray = dcm.pixel_array
+
+            # Just take first channel if it's RGB?
+            if len(pixelArray.shape) == 4:
+                pixelArray = pixelArray[:,:,:,0]
+
+            n_slices = pixelArray.shape[0]
+
+            slice_ix = 0
+        except:
+            selectedDicom = None
+            # continue
+
+        with tab1:
+
+            with stepOne:
+                st.write("Select frame for annotation. Aim for an end-diastolic frame with good visualisation of the artery of interest.")
+
+                slice_ix = st.slider('Frame', 0, n_slices-1, int(n_slices/2), key='sliceSlider')
+
+
+                predictedMask = numpy.zeros_like(pixelArray[slice_ix, :, :])
+
+
+            with stepTwo:
+
+                selectedArtery = st.selectbox("Select artery for annotation:",
+                        ['LAD', 'CX', 'RCA', 'LM', 'OM', 'AM', 'D'],
+                        key="arteryDropMenu"
+                    )
+
+                st.write("Beginning with the desired start point and finishing at the desired end point, click along the artery aiming for ~5-10 points.")
+
+
+                stroke_color = angioPyFunctions.colourTableList[selectedArtery]
+
+
+            col1, col2 = st.columns((15,15))
+
+            with col1:
+                col1a, col1b, col1c = st.columns((1,10,1))
+
+                with col1b:
+
+                    leftImageText = "<p style='text-align: center; color: white;'>Beginning with the desired <u><b>start point</b></u> and finishing at the desired <u><b>end point</b></u>, click along the artery aiming for ~5-10 points. Segmentation is automatic.</p>"
+
+                    st.markdown(f"<h5 style='text-align: center; color: white;'>Selected frame</h5>", unsafe_allow_html=True)
+
+                    st.markdown(leftImageText, unsafe_allow_html=True)
+
+                    selectedFrame = pixelArray[slice_ix, :, :]
+                    selectedFrame = cv2.resize(selectedFrame, (512,512))
+
+                    # Create a canvas component
+                    annotationCanvas = st_canvas(
+                        fill_color="red",  # Fixed fill color with some opacity
+                        stroke_width=1,
+                        stroke_color="red",
+                        background_color='black',
+                        background_image= Image.fromarray(selectedFrame),
+                        update_streamlit=True,
+                        height=512,
+                        width=512,
+                        drawing_mode="point",
+                        point_display_radius=2,
+                        key=st.session_state.dicomDropDown,
+                    )
+
+
+                    # Do something interesting with the image data and paths
+                    if annotationCanvas.json_data is not None:
+                        objects = pd.json_normalize(annotationCanvas.json_data["objects"]) # need to convert obj to str because PyArrow
+
+                        if len(objects) != 0:
+
+                            for col in objects.select_dtypes(include=['object']).columns:
+                                objects[col] = objects[col].astype("str")
+
+                            groundTruthPoints = numpy.vstack(
+                                (
+                                    numpy.array(objects['top']),
+                                    numpy.array(objects['left']+3.5) # compensate for some streamlit offset or something
+                                )
+                            ).T
+
+                            mask = angioPyFunctions.arterySegmentation(
+                                pixelArray[slice_ix],
+                                groundTruthPoints,
+                            )
+                            predictedMask = predict.CoronaryDataset.mask2image(mask)
+                            # predictedMask = predictedMask.crop((0, 0, imageSize[0], imageSize[1]))
+                            predictedMask = numpy.asarray(predictedMask)
+
+            with col2:
+                col2a, col2b, col2c = st.columns((1,10,1))
+
+                with col2b:
+                    st.markdown(f"<h5 style='text-align: center; color: white;'>Predicted mask</h1>", unsafe_allow_html=True)
+                    st.markdown(f"<p style='text-align: center; color: white;'>If the predicted mask has errors, restart and select more points to help the segmentation model. </p>", unsafe_allow_html=True)
+
+                    stroke_color = "rgba(255, 255, 255, 255)"
+
+                    maskCanvas = st_canvas(
+                        fill_color=angioPyFunctions.colourTableList[selectedArtery],  # Fixed fill color with some opacity
+                        stroke_width=0,
+                        stroke_color=stroke_color,
+                        background_color='black',
+                        background_image= Image.fromarray(predictedMask),
+                        update_streamlit=True,
+                        height=512,
+                        width=512,
+                        drawing_mode="freedraw",
+                        point_display_radius=3,
+                        key="maskCanvas",
+                    )
+
+
+                    # Check that the mask array is not blank
+                    if numpy.sum(predictedMask) > 0 and len(objects)>4:
+                        # add alpha channel to predict mask in order to merge
+                        b_channel, g_channel, r_channel = cv2.split(predictedMask)
+                        a_channel = numpy.full_like(predictedMask[:,:,0], fill_value=255)
+
+                        predictedMaskRGBA = cv2.merge((predictedMask, a_channel))
+
+
+                        with tab2:
+                            # combinedMask = cv2.cvtColor(predictedMaskRGBA, cv2.COLOR_RGBA2RGB)
+
+                            # print(combinedMask.shape)
+                            # tifffile.imwrite(f"{outputPath}/test.tif", combinedMask)
+
+
+                            # tab2Col1, tab2Col2, tab2Col3 = st.columns([1,15,1])
+                            tab2Col1, tab2Col2 = st.columns([20,10])
+
+                            with tab2Col1:
+                                st.markdown(f"<h5 style='text-align: center; color: white;'><br>Artery profile</h5>", unsafe_allow_html=True)
+
+                                # Extract thickness information from mask
+                                EDT = scipy.ndimage.distance_transform_edt(cv2.cvtColor(predictedMaskRGBA, cv2.COLOR_RGBA2GRAY))
+
+                                # Skeletonise, get a list of ordered centreline points, and spline them
+                                skel = angioPyFunctions.skeletonise(predictedMaskRGBA)
+                                tck = angioPyFunctions.skelSplinerWithThickness(skel=skel, EDT=EDT)
+
+                                # Interogate the spline function over 1000 points
+                                splinePointsY, splinePointsX, splineThicknesses = scipy.interpolate.splev(
+                                numpy.linspace(
+                                    0.0,
+                                    1.0,
+                                    1000), 
+                                    tck)
+
+                                clippingLength = 20
+
+                                vesselThicknesses = splineThicknesses[clippingLength:-clippingLength]*2
+
+                                fig = px.line(x=numpy.arange(1,len(vesselThicknesses)+1),y=vesselThicknesses, labels=dict(x="Centreline point", y="Thickness (pixels)"), width=800)
+                                # fig.update_layout(showlegend=False, xaxis={'showgrid': False, 'zeroline': True})
+                                fig.update_traces(line_color='rgb(31, 119, 180)', textfont_color="white", line={'width':4})
+                                fig.update_xaxes(showline=True, linewidth=2, linecolor='white', showgrid=False,gridcolor='white')
+                                fig.update_yaxes(showline=True, linewidth=2, linecolor='white', gridcolor='white')
+
+                                fig.update_layout(yaxis_range=[0,numpy.max(vesselThicknesses)*1.2])
+                                fig.update_layout(font_color="white",title_font_color="white")
+                                fig.update_layout({'plot_bgcolor': 'rgba(0, 0, 0, 0)','paper_bgcolor': 'rgba(0, 0, 0, 0)'})
+
+
+                                selected_points = plotly_events(fig)
+
+
+
+                            with tab2Col2:
+
+                                st.markdown(f"<h5 style='text-align: center; color: white;'><br>Contours</h5>", unsafe_allow_html=True)
+
+
+                                selectedFrameRGBA = cv2.cvtColor(selectedFrame, cv2.COLOR_GRAY2RGBA)
+
+                                contour = angioPyFunctions.maskOutliner(labelledArtery=predictedMaskRGBA[:,:,0], outlineThickness=1)
+
+                                selectedFrameRGBA[contour, :] =    [angioPyFunctions.colourTableList[selectedArtery][2],
+                                                                    angioPyFunctions.colourTableList[selectedArtery][1],
+                                                                    angioPyFunctions.colourTableList[selectedArtery][0],
+                                                                    255]
+
+                                fig2 = px.imshow(selectedFrameRGBA)
+
+
+                                fig2.update_xaxes(visible=False)
+                                fig2.update_yaxes(visible=False)
+                                fig2.update_layout(margin={"t": 0, "b": 0, "r": 0, "l": 0, "pad": 0},) #remove margins
+                                # fig2.coloraxis(visible=False)
+
+                                fig2.update_traces(dict(
+                                    showscale=False, 
+                                    coloraxis=None, 
+                                    colorscale='gray'), selector={'type':'heatmap'})
+
+                                fig2.add_trace(go.Scatter(x=splinePointsX[clippingLength:-clippingLength], y=splinePointsY[clippingLength:-clippingLength], line=dict(width=1)))
+
+                                st.plotly_chart(fig2, use_container_width=True)

predict.py

@@ -0,0 +1,65 @@
+import argparse
+import logging
+import os
+
+import torch
+import torch.nn as nn
+from PIL import Image
+from torchvision import transforms
+
+from utils.dataset import CoronaryDataset
+import segmentation_models_pytorch.segmentation_models_pytorch as smp
+
+from torch.backends import cudnn
+
+'''
+This uses a pytorch coronary segmentation model (EfficientNetPLusPlus) that has been trained using a freely available dataset of labelled coronary angiograms from: http://personal.cimat.mx:8181/~ivan.cruz/DB_Angiograms.html
+The input is a raw angiogram image, and the output is a segmentation mask of all the arteries. This output will be used as the 'first guess' to speed up artery annotation. 
+'''
+
+def predict_img(net, dataset_class, full_img, device, scale_factor=1, n_classes=3):
+    # NOTE n_classes is the number of possible values that can be predicted for a given pixel. In a standard binary segmentation task, this will be 2 i.e. black or white
+
+    net.eval()
+
+    img = torch.from_numpy(dataset_class.preprocess(full_img, scale_factor))
+
+    img = img.unsqueeze(0)
+    img = img.to(device=device, dtype=torch.float32)
+
+    with torch.no_grad():
+        output = net(img)
+
+        if n_classes > 1:
+            probs = torch.softmax(output, dim=1)
+        else:
+            probs = torch.sigmoid(output)
+
+        probs = probs.squeeze(0)
+
+        tf = transforms.Compose(
+            [
+                transforms.ToPILImage(),
+                transforms.Resize(full_img.size[1]),
+                transforms.ToTensor()
+            ]
+        )
+
+        full_mask = tf(probs.cpu())   
+
+    if n_classes > 1:
+        return dataset_class.one_hot2mask(full_mask)
+    else:
+        return full_mask > 0.5
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='Predict masks from input images', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+    # parser.add_argument('-d', '--dataset', type=str, help='Specifies the dataset to be used', dest='dataset', required=True)
+    parser.add_argument('--model', '-m', default='MODEL.pth', metavar='FILE', help="Specify the file in which the model is stored")
+    parser.add_argument('--input', '-i', metavar='INPUT', nargs='+', help='filenames of input images', required=True)
+    parser.add_argument('--output', '-o', metavar='INPUT', nargs='+', help='Filenames of output images')
+
+    return parser.parse_args()
+
+

requirements.txt

@@ -0,0 +1,29 @@
+# Automatically generated by https://github.com/damnever/pigar.
+
+astropy==5.2.2
+efficientnet-pytorch==0.7.1
+fil-finder==1.7.2
+matplotlib==3.7.2
+mpld3==0.5.9
+numpy==1.24.4
+opencv-python==4.8.0.76
+pandas==2.0.3
+Pillow==9.5.0
+plotly==5.16.1
+pretrainedmodels==0.7.4
+pydicom==2.4.3
+PyYAML==6.0.1
+scikit-image==0.21.0
+scikit-learn==1.3.0
+scipy==1.10.1
+setuptools==47.1.0
+SimpleITK==2.2.1
+streamlit<=1.38.0
+streamlit-drawable-canvas==0.9.3
+streamlit-plotly-events==0.0.6
+tifffile==2023.7.10
+timm==0.9.6
+torch==2.0.1
+torchvision==0.15.2
+tqdm==4.61.1
+pooch

segmentation_models_pytorch/.github/FUNDING.yml

@@ -0,0 +1,12 @@
+# These are supported funding model platforms
+
+github: # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: qubvel
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: qubvel
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

segmentation_models_pytorch/.github/stale.yml

@@ -0,0 +1,17 @@
+# Number of days of inactivity before an issue becomes stale
+daysUntilStale: 60
+# Number of days of inactivity before a stale issue is closed
+daysUntilClose: 7
+# Issues with these labels will never be considered stale
+exemptLabels:
+  - pinned
+  - security
+# Label to use when marking an issue as stale
+staleLabel: wontfix
+# Comment to post when marking an issue as stale. Set to `false` to disable
+markComment: >
+  This issue has been automatically marked as stale because it has not had
+  recent activity. It will be closed if no further activity occurs. Thank you
+  for your contributions.
+# Comment to post when closing a stale issue. Set to `false` to disable
+closeComment: false

segmentation_models_pytorch/.github/workflows/pypi.yml

@@ -0,0 +1,26 @@
+name: Upload Python Package
+
+on:
+  release:
+    types: [published]
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python
+      uses: actions/setup-python@v2
+      with:
+        python-version: '3.6'
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install setuptools wheel twine mock
+    - name: Build and publish
+      env:
+        TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
+        TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
+      run: |
+        python setup.py sdist bdist_wheel
+        twine upload dist/*

segmentation_models_pytorch/.github/workflows/tests.yml

@@ -0,0 +1,34 @@
+
+# This workflow will install Python dependencies, run tests and lint with a variety of Python versions
+# For more information see: https://help.github.com/actions/language-and-framework-guides/using-python-with-github-actions
+
+name: CI
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  test:
+
+    runs-on: ubuntu-18.04
+
+    steps:
+    - uses: actions/checkout@v2
+    
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.6
+        
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install codecov pytest mock
+        pip install torch==1.7.1+cpu torchvision==0.8.2+cpu torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
+        pip install .
+    - name: Test
+      run: |
+        python -m pytest -s tests

segmentation_models_pytorch/.gitignore

@@ -0,0 +1,105 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+.idea/
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/

segmentation_models_pytorch/HALLOFFAME.md

@@ -0,0 +1,90 @@
+# Hall of Fame
+
+`Segmentation Models` package is widely used in the image segmentation competitions.
+Here you can find competitions, names of the winners and links to their solutions.
+
+Please, follow these rules, when adding a solution to the "Hall of Fame":
+
+1. Solution should be high rated (e.g. for Kaggle gold or silver medal) 
+2. There should be a description of the solution (post at the forum / code / blog post / paper / pre-print)
+
+
+## Kaggle
+
+### [Severstal: Steel Defect Detection](https://www.kaggle.com/c/severstal-steel-defect-detection)
+
+- 1st place. 
+[Wuxi Jiangsu](https://www.kaggle.com/rguo97), 
+[Hongbo Zhu](https://www.kaggle.com/zhuhongbo), 
+[Yizhuo Yu](https://www.kaggle.com/paffpaffyu)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114254#latest-675874)]
+
+- 5th place. 
+[Guanshuo Xu](https://www.kaggle.com/wowfattie)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/117208#latest-675385)]
+
+- 9th place. 
+[Jacek Poplawski](https://www.linkedin.com/in/jacekpoplawski/)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114297#latest-660842)]
+
+- 10th place.
+[Alexey Rozhkov](https://www.linkedin.com/in/alexisrozhkov)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114465#latest-659615)]
+
+- 12th place. 
+[Pavel Yakubovskiy](https://www.linkedin.com/in/pavel-yakubovskiy/), 
+[Ilya Dobrynin](https://www.linkedin.com/in/ilya-dobrynin-79a89b106/), 
+[Denis Kolpakov](https://www.linkedin.com/in/denis-kolpakov-ab3137197/)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114309#latest-661404)]
+
+- 31st place. 
+[Insaf Ashrapov](https://www.linkedin.com/in/iashrapov/), 
+[Igor Krashenyi](https://www.linkedin.com/in/igor-krashenyi-38b89b98), 
+[Pavel Pleskov](https://www.linkedin.com/in/ppleskov), 
+[Anton Zakharenkov](https://www.linkedin.com/in/anton-zakharenkov/), 
+[Nikolai Popov](https://www.linkedin.com/in/nikolai-popov-b2157370/)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114383#latest-658438)]
+[[code](https://github.com/Diyago/Severstal-Steel-Defect-Detection)]
+
+- 55th place. 
+[Karl Hornlund](https://www.linkedin.com/in/karl-hornlund/)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/114410#latest-672682)]
+[[code](https://github.com/khornlund/severstal-steel-defect-detection)]
+
+- Efficiency round 1st place.
+[Stefan Stefanov](https://www.linkedin.com/in/stefan-stefanov-63a77b1)  
+[[description](https://www.kaggle.com/c/severstal-steel-defect-detection/discussion/117486#latest-674229)]
+
+
+### [Understanding Clouds from Satellite Images](https://www.kaggle.com/c/understanding_cloud_organization)
+
+- 2nd place.
+[Andrey Kiryasov](https://www.kaggle.com/ekydna)  
+[[description](https://www.kaggle.com/c/understanding_cloud_organization/discussion/118255#latest-678189)]
+
+- 4th place.
+[Ching-Loong Seow](https://www.linkedin.com/in/clseow/)  
+[[description](https://www.kaggle.com/c/understanding_cloud_organization/discussion/118016#latest-677333)]
+
+- 34th place.
+[Karl Hornlund](https://www.linkedin.com/in/karl-hornlund/)  
+[[description](https://www.kaggle.com/c/understanding_cloud_organization/discussion/118250#latest-678176)]
+[[code](https://github.com/khornlund/understanding-cloud-organization)]
+
+- 55th place.
+[Pavel Yakubovskiy](https://www.linkedin.com/in/pavel-yakubovskiy/)  
+[[description](https://www.kaggle.com/c/understanding_cloud_organization/discussion/118019#latest-678626)]
+
+## Other platforms
+
+### [MICCAI 2020 TN-SCUI challenge](https://tn-scui2020.grand-challenge.org/Home/)  
+- 1st place.
+[Mingyu Wang](https://github.com/WAMAWAMA)  
+[[description](https://github.com/WAMAWAMA/TNSCUI2020-Seg-Rank1st)]
+[[code](https://github.com/WAMAWAMA/TNSCUI2020-Seg-Rank1st)]
+
+### [Open Cities AI Challenge: Segmenting Buildings for Disaster Resilience](https://www.drivendata.org/competitions/60/building-segmentation-disaster-resilience/)
+ - 1st place.
+[Pavel Yakubovskiy](https://www.linkedin.com/in/pavel-yakubovskiy/).  
+[[code and description](https://github.com/qubvel/open-cities-challenge)]
+

segmentation_models_pytorch/LICENSE

@@ -0,0 +1,21 @@
+The MIT License
+
+Copyright (c) 2019, Pavel Yakubovskiy
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

segmentation_models_pytorch/MANIFEST.in

@@ -0,0 +1 @@
+include README.md LICENSE requirements.txt

segmentation_models_pytorch/README.md

@@ -0,0 +1,409 @@
+<div align="center">
+ 
+![logo](https://i.ibb.co/dc1XdhT/Segmentation-Models-V2-Side-1-1.png)  
+**Python library with Neural Networks for Image  
+Segmentation based on [PyTorch](https://pytorch.org/).**  
+
+[![Documentation Status](https://readthedocs.org/projects/smp/badge/?version=latest)](https://segmentation-models-pytorch.readthedocs.io/en/latest/?badge=latest) <br> [![Generic badge](https://img.shields.io/badge/License-MIT-<COLOR>.svg)](https://shields.io/)
+
+</div>
+
+The main features of this library are:
+
+ - High level API (just two lines to create a neural network)
+ - 12 models architectures for binary and multi class segmentation (including legendary Unet)
+ - 104 available encoders
+ - All encoders have pre-trained weights for faster and better convergence
+ 
+### [📚 Project Documentation 📚](http://smp.readthedocs.io/)
+
+Visit [Read The Docs Project Page](https://segmentation-models-pytorch.readthedocs.io/en/latest/) or read following README to know more about Segmentation Models Pytorch (SMP for short) library
+
+### 📋 Table of content
+ 1. [Quick start](#start)
+ 2. [Examples](#examples)
+ 3. [Models](#models)
+    1. [Architectures](#architectures)
+    2. [Encoders](#encoders)
+ 4. [Models API](#api)
+    1. [Input channels](#input-channels)
+    2. [Auxiliary classification output](#auxiliary-classification-output)
+    3. [Depth](#depth)
+ 5. [Installation](#installation)
+ 6. [Competitions won with the library](#competitions-won-with-the-library)
+ 7. [Contributing](#contributing)
+ 8. [Citing](#citing)
+ 9. [License](#license)
+
+### ⏳ Quick start <a name="start"></a>
+
+#### 1. Create your first Segmentation model with SMP
+
+Segmentation model is just a PyTorch nn.Module, which can be created as easy as:
+
+```python
+import segmentation_models_pytorch as smp
+
+model = smp.Unet(
+    encoder_name="resnet34",        # choose encoder, e.g. mobilenet_v2 or efficientnet-b7
+    encoder_weights="imagenet",     # use `imagenet` pre-trained weights for encoder initialization
+    in_channels=1,                  # model input channels (1 for gray-scale images, 3 for RGB, etc.)
+    classes=3,                      # model output channels (number of classes in your dataset)
+)
+```
+ - see [table](#architectures) with available model architectures
+ - see [table](#encoders) with available encoders and their corresponding weights
+
+#### 2. Configure data preprocessing
+
+All encoders have pretrained weights. Preparing your data the same way as during weights pre-training may give your better results (higher metric score and faster convergence). But it is relevant only for 1-2-3-channels images and **not necessary** in case you train the whole model, not only decoder.
+
+```python
+from segmentation_models_pytorch.encoders import get_preprocessing_fn
+
+preprocess_input = get_preprocessing_fn('resnet18', pretrained='imagenet')
+```
+
+Congratulations! You are done! Now you can train your model with your favorite framework!
+
+### 💡 Examples <a name="examples"></a>
+ - Training model for cars segmentation on CamVid dataset [here](https://github.com/qubvel/segmentation_models.pytorch/blob/master/examples/cars%20segmentation%20(camvid).ipynb).
+ - Training SMP model with [Catalyst](https://github.com/catalyst-team/catalyst) (high-level framework for PyTorch), [TTAch](https://github.com/qubvel/ttach) (TTA library for PyTorch) and [Albumentations](https://github.com/albu/albumentations) (fast image augmentation library) - [here](https://github.com/catalyst-team/catalyst/blob/master/examples/notebooks/segmentation-tutorial.ipynb) [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/catalyst-team/catalyst/blob/master/examples/notebooks/segmentation-tutorial.ipynb)
+ - Training SMP model with [Pytorch-Lightning](https://pytorch-lightning.readthedocs.io) framework - [here](https://github.com/ternaus/cloths_segmentation) (clothes binary segmentation by [@teranus](https://github.com/ternaus)).
+
+### 📦 Models <a name="models"></a>
+
+#### Architectures <a name="architectures"></a>
+ - Unet [[paper](https://arxiv.org/abs/1505.04597)] [[docs](https://smp.readthedocs.io/en/latest/models.html#unet)]
+ - Unet++ [[paper1](https://arxiv.org/abs/1807.10165), [paper2](https://arxiv.org/abs/1912.05074)] [[docs](https://smp.readthedocs.io/en/latest/models.html#id2)]
+ - EfficientUNet++ [[paper]()] [[docs](https://segmentation-models-pytorch.readthedocs.io/en/latest/models.html#efficientunet)]
+ - ResUnet [[paper](https://arxiv.org/abs/1711.10684)] [[docs](https://segmentation-models-pytorch.readthedocs.io/en/latest/models.html#resunet)]
+ - ResUnet++ [[paper](https://arxiv.org/abs/1911.07067)] [[docs](https://segmentation-models-pytorch.readthedocs.io/en/latest/models.html#id4)]
+ - MAnet [[paper](https://ieeexplore.ieee.org/abstract/document/9201310)] [[docs](https://smp.readthedocs.io/en/latest/models.html#manet)]
+ - Linknet [[paper](https://arxiv.org/abs/1707.03718)] [[docs](https://smp.readthedocs.io/en/latest/models.html#linknet)]
+ - FPN [[paper](http://presentations.cocodataset.org/COCO17-Stuff-FAIR.pdf)] [[docs](https://smp.readthedocs.io/en/latest/models.html#fpn)]
+ - PSPNet [[paper](https://arxiv.org/abs/1612.01105)] [[docs](https://smp.readthedocs.io/en/latest/models.html#pspnet)]
+ - PAN [[paper](https://arxiv.org/abs/1805.10180)] [[docs](https://smp.readthedocs.io/en/latest/models.html#pan)]
+ - DeepLabV3 [[paper](https://arxiv.org/abs/1706.05587)] [[docs](https://smp.readthedocs.io/en/latest/models.html#deeplabv3)]
+ - DeepLabV3+ [[paper](https://arxiv.org/abs/1802.02611)] [[docs](https://smp.readthedocs.io/en/latest/models.html#id9)]
+
+#### Encoders <a name="encoders"></a>
+
+The following is a list of supported encoders in the SMP. Select the appropriate family of encoders and click to expand the table and select a specific encoder and its pre-trained weights (`encoder_name` and `encoder_weights` parameters).
+
+<details>
+<summary style="margin-left: 25px;">ResNet</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|resnet18                        |imagenet / ssl / swsl           |11M                             |
+|resnet34                        |imagenet                        |21M                             |
+|resnet50                        |imagenet / ssl / swsl           |23M                             |
+|resnet101                       |imagenet                        |42M                             |
+|resnet152                       |imagenet                        |58M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">ResNeXt</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|resnext50_32x4d                 |imagenet / ssl / swsl           |22M                             |
+|resnext101_32x4d                |ssl / swsl                      |42M                             |
+|resnext101_32x8d                |imagenet / instagram / ssl / swsl|86M                         |
+|resnext101_32x16d               |instagram / ssl / swsl          |191M                            |
+|resnext101_32x32d               |instagram                       |466M                            |
+|resnext101_32x48d               |instagram                       |826M                            |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">ResNeSt</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|timm-resnest14d                 |imagenet                        |8M                              |
+|timm-resnest26d                 |imagenet                        |15M                             |
+|timm-resnest50d                 |imagenet                        |25M                             |
+|timm-resnest101e                |imagenet                        |46M                             |
+|timm-resnest200e                |imagenet                        |68M                             |
+|timm-resnest269e                |imagenet                        |108M                            |
+|timm-resnest50d_4s2x40d         |imagenet                        |28M                             |
+|timm-resnest50d_1s4x24d         |imagenet                        |23M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">Res2Ne(X)t</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|timm-res2net50_26w_4s           |imagenet                        |23M                             |
+|timm-res2net101_26w_4s          |imagenet                        |43M                             |
+|timm-res2net50_26w_6s           |imagenet                        |35M                             |
+|timm-res2net50_26w_8s           |imagenet                        |46M                             |
+|timm-res2net50_48w_2s           |imagenet                        |23M                             |
+|timm-res2net50_14w_8s           |imagenet                        |23M                             |
+|timm-res2next50                 |imagenet                        |22M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">RegNet(x/y)</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|timm-regnetx_002                |imagenet                        |2M                              |
+|timm-regnetx_004                |imagenet                        |4M                              |
+|timm-regnetx_006                |imagenet                        |5M                              |
+|timm-regnetx_008                |imagenet                        |6M                              |
+|timm-regnetx_016                |imagenet                        |8M                              |
+|timm-regnetx_032                |imagenet                        |14M                             |
+|timm-regnetx_040                |imagenet                        |20M                             |
+|timm-regnetx_064                |imagenet                        |24M                             |
+|timm-regnetx_080                |imagenet                        |37M                             |
+|timm-regnetx_120                |imagenet                        |43M                             |
+|timm-regnetx_160                |imagenet                        |52M                             |
+|timm-regnetx_320                |imagenet                        |105M                            |
+|timm-regnety_002                |imagenet                        |2M                              |
+|timm-regnety_004                |imagenet                        |3M                              |
+|timm-regnety_006                |imagenet                        |5M                              |
+|timm-regnety_008                |imagenet                        |5M                              |
+|timm-regnety_016                |imagenet                        |10M                             |
+|timm-regnety_032                |imagenet                        |17M                             |
+|timm-regnety_040                |imagenet                        |19M                             |
+|timm-regnety_064                |imagenet                        |29M                             |
+|timm-regnety_080                |imagenet                        |37M                             |
+|timm-regnety_120                |imagenet                        |49M                             |
+|timm-regnety_160                |imagenet                        |80M                             |
+|timm-regnety_320                |imagenet                        |141M                            |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">SE-Net</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|senet154                        |imagenet                        |113M                            |
+|se_resnet50                     |imagenet                        |26M                             |
+|se_resnet101                    |imagenet                        |47M                             |
+|se_resnet152                    |imagenet                        |64M                             |
+|se_resnext50_32x4d              |imagenet                        |25M                             |
+|se_resnext101_32x4d             |imagenet                        |46M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">SK-ResNe(X)t</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|timm-skresnet18                 |imagenet                        |11M                             |
+|timm-skresnet34                 |imagenet                        |21M                             |
+|timm-skresnext50_32x4d          |imagenet                        |25M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">DenseNet</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|densenet121                     |imagenet                        |6M                              |
+|densenet169                     |imagenet                        |12M                             |
+|densenet201                     |imagenet                        |18M                             |
+|densenet161                     |imagenet                        |26M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">Inception</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|inceptionresnetv2               |imagenet /  imagenet+background |54M                             |
+|inceptionv4                     |imagenet /  imagenet+background |41M                             |
+|xception                        |imagenet                        |22M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">EfficientNet</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|efficientnet-b0                 |imagenet                        |4M                              |
+|efficientnet-b1                 |imagenet                        |6M                              |
+|efficientnet-b2                 |imagenet                        |7M                              |
+|efficientnet-b3                 |imagenet                        |10M                             |
+|efficientnet-b4                 |imagenet                        |17M                             |
+|efficientnet-b5                 |imagenet                        |28M                             |
+|efficientnet-b6                 |imagenet                        |40M                             |
+|efficientnet-b7                 |imagenet                        |63M                             |
+|timm-efficientnet-b0            |imagenet / advprop / noisy-student|4M                              |
+|timm-efficientnet-b1            |imagenet / advprop / noisy-student|6M                              |
+|timm-efficientnet-b2            |imagenet / advprop / noisy-student|7M                              |
+|timm-efficientnet-b3            |imagenet / advprop / noisy-student|10M                             |
+|timm-efficientnet-b4            |imagenet / advprop / noisy-student|17M                             |
+|timm-efficientnet-b5            |imagenet / advprop / noisy-student|28M                             |
+|timm-efficientnet-b6            |imagenet / advprop / noisy-student|40M                             |
+|timm-efficientnet-b7            |imagenet / advprop / noisy-student|63M                             |
+|timm-efficientnet-b8            |imagenet / advprop             |84M                             |
+|timm-efficientnet-l2            |noisy-student                   |474M                            |
+|timm-efficientnet-lite0         |imagenet                        |4M                              |
+|timm-efficientnet-lite1         |imagenet                        |5M                              |
+|timm-efficientnet-lite2         |imagenet                        |6M                              |
+|timm-efficientnet-lite3         |imagenet                        |8M                             |
+|timm-efficientnet-lite4         |imagenet                        |13M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">MobileNet</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|mobilenet_v2                    |imagenet                        |2M                              |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">DPN</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|dpn68                           |imagenet                        |11M                             |
+|dpn68b                          |imagenet+5k                     |11M                             |
+|dpn92                           |imagenet+5k                     |34M                             |
+|dpn98                           |imagenet                        |58M                             |
+|dpn107                          |imagenet+5k                     |84M                             |
+|dpn131                          |imagenet                        |76M                             |
+
+</div>
+</details>
+
+<details>
+<summary style="margin-left: 25px;">VGG</summary>
+<div style="margin-left: 25px;">
+
+|Encoder                         |Weights                         |Params, M                       |
+|--------------------------------|:------------------------------:|:------------------------------:|
+|vgg11                           |imagenet                        |9M                              |
+|vgg11_bn                        |imagenet                        |9M                              |
+|vgg13                           |imagenet                        |9M                              |
+|vgg13_bn                        |imagenet                        |9M                              |
+|vgg16                           |imagenet                        |14M                             |
+|vgg16_bn                        |imagenet                        |14M                             |
+|vgg19                           |imagenet                        |20M                             |
+|vgg19_bn                        |imagenet                        |20M                             |
+
+</div>
+</details>
+
+
+\* `ssl`, `swsl` - semi-supervised and weakly-supervised learning on ImageNet ([repo](https://github.com/facebookresearch/semi-supervised-ImageNet1K-models)).
+
+
+### 🔁 Models API <a name="api"></a>
+
+ - `model.encoder` - pretrained backbone to extract features of different spatial resolution
+ - `model.decoder` - depends on models architecture (`Unet`/`Linknet`/`PSPNet`/`FPN`)
+ - `model.segmentation_head` - last block to produce required number of mask channels (include also optional upsampling and activation)
+ - `model.classification_head` - optional block which create classification head on top of encoder
+ - `model.forward(x)` - sequentially pass `x` through model\`s encoder, decoder and segmentation head (and classification head if specified)
+
+##### Input channels
+Input channels parameter allows you to create models, which process tensors with arbitrary number of channels.
+If you use pretrained weights from imagenet - weights of first convolution will be reused for
+1- or 2- channels inputs, for input channels > 4 weights of first convolution will be initialized randomly.
+```python
+model = smp.FPN('resnet34', in_channels=1)
+mask = model(torch.ones([1, 1, 64, 64]))
+```
+
+##### Auxiliary classification output  
+All models support `aux_params` parameters, which is default set to `None`. 
+If `aux_params = None` then classification auxiliary output is not created, else
+model produce not only `mask`, but also `label` output with shape `NC`.
+Classification head consists of GlobalPooling->Dropout(optional)->Linear->Activation(optional) layers, which can be 
+configured by `aux_params` as follows:
+```python
+aux_params=dict(
+    pooling='avg',             # one of 'avg', 'max'
+    dropout=0.5,               # dropout ratio, default is None
+    activation='sigmoid',      # activation function, default is None
+    classes=4,                 # define number of output labels
+)
+model = smp.Unet('resnet34', classes=4, aux_params=aux_params)
+mask, label = model(x)
+```
+
+##### Depth
+Depth parameter specify a number of downsampling operations in encoder, so you can make
+your model lighter if specify smaller `depth`.
+```python
+model = smp.Unet('resnet34', encoder_depth=4)
+```
+
+
+### 🛠 Installation <a name="installation"></a>
+Latest version from source:
+```bash
+$ pip install git+https://github.com/jlcsilva/segmentation_models.pytorch
+````
+
+### 🏆 Competitions won with the library
+
+`Segmentation Models` package is widely used in the image segmentation competitions.
+[Here](https://github.com/qubvel/segmentation_models.pytorch/blob/master/HALLOFFAME.md) you can find competitions, names of the winners and links to their solutions.
+
+### 🤝 Contributing
+
+##### Run test
+```bash
+$ docker build -f docker/Dockerfile.dev -t smp:dev . && docker run --rm smp:dev pytest -p no:cacheprovider
+```
+##### Generate table
+```bash
+$ docker build -f docker/Dockerfile.dev -t smp:dev . && docker run --rm smp:dev python misc/generate_table.py
+```
+
+### 📝 Citing
+```
+@misc{Yakubovskiy:2019,
+  Author = {Pavel Yakubovskiy},
+  Title = {Segmentation Models Pytorch},
+  Year = {2020},
+  Publisher = {GitHub},
+  Journal = {GitHub repository},
+  Howpublished = {\url{https://github.com/qubvel/segmentation_models.pytorch}}
+}
+```
+
+### 🛡️ License <a name="license"></a>
+Project is distributed under [MIT License](https://github.com/qubvel/segmentation_models.pytorch/blob/master/LICENSE)

segmentation_models_pytorch/__init__.py

@@ -0,0 +1 @@
+from segmentation_models_pytorch import *

segmentation_models_pytorch/docker/Dockerfile

@@ -0,0 +1,3 @@
+FROM anibali/pytorch:cuda-9.0
+
+RUN pip install segmentation-models-pytorch

segmentation_models_pytorch/docker/Dockerfile.dev

@@ -0,0 +1,10 @@
+FROM anibali/pytorch:1.5.0-nocuda
+
+WORKDIR /tmp/smp/
+
+COPY ./requirements.txt /tmp/smp/requirements.txt
+RUN pip install -r requirements.txt
+RUN pip install pytest mock
+
+COPY . /tmp/smp/
+RUN pip install .

segmentation_models_pytorch/docs/Makefile

@@ -0,0 +1,20 @@
+# Minimal makefile for Sphinx documentation
+#
+
+# You can set these variables from the command line, and also
+# from the environment for the first two.
+SPHINXOPTS    ?=
+SPHINXBUILD   ?= sphinx-build
+SOURCEDIR     = .
+BUILDDIR      = build
+
+# Put it first so that "make" without argument is like "make help".
+help:
+	@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+.PHONY: help Makefile
+
+# Catch-all target: route all unknown targets to Sphinx using the new
+# "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
+%: Makefile
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)

segmentation_models_pytorch/docs/conf.py

@@ -0,0 +1,120 @@
+# Configuration file for the Sphinx documentation builder.
+#
+# This file only contains a selection of the most common options. For a full
+# list see the documentation:
+# https://www.sphinx-doc.org/en/master/usage/configuration.html
+
+# -- Path setup --------------------------------------------------------------
+
+# If extensions (or modules to document with autodoc) are in another directory,
+# add these directories to sys.path here. If the directory is relative to the
+# documentation root, use os.path.abspath to make it absolute, like shown here.
+#
+# import os
+# import sys
+# sys.path.insert(0, os.path.abspath('.'))
+
+import os
+import re
+import sys
+import datetime
+sys.path.append('..')
+
+# -- Project information -----------------------------------------------------
+
+project = 'Segmentation Models'
+copyright = '{}, Pavel Yakubovskiy'.format(datetime.datetime.now().year)
+author = 'Pavel Yakubovskiy'
+
+def get_version():
+    sys.path.append('../segmentation_models_pytorch')
+    from __version__ import __version__ as version
+    sys.path.pop(-1)
+    return version
+
+version = get_version()
+
+# -- General configuration ---------------------------------------------------
+
+# Add any Sphinx extension module names here, as strings. They can be
+# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
+# ones.
+
+extensions = [
+    'sphinx.ext.autodoc',
+    'sphinx.ext.coverage',
+    'sphinx.ext.napoleon',
+    'sphinx.ext.viewcode',
+    'sphinx.ext.mathjax', 
+]
+
+# Add any paths that contain templates here, relative to this directory.
+templates_path = ['_templates']
+
+# List of patterns, relative to source directory, that match files and
+# directories to ignore when looking for source files.
+# This pattern also affects html_static_path and html_extra_path.
+exclude_patterns = []
+
+
+# -- Options for HTML output -------------------------------------------------
+
+# The theme to use for HTML and HTML Help pages.  See the documentation for
+# a list of builtin themes.
+#
+
+import sphinx_rtd_theme
+html_theme = "sphinx_rtd_theme"
+html_theme_path = [sphinx_rtd_theme.get_html_theme_path()]
+
+# import karma_sphinx_theme
+# html_theme = "karma_sphinx_theme"
+import faculty_sphinx_theme
+html_theme = "faculty_sphinx_theme"
+
+# import catalyst_sphinx_theme
+# html_theme = "catalyst_sphinx_theme"
+# html_theme_path = [catalyst_sphinx_theme.get_html_theme_path()]
+
+html_logo = "logo.png"
+
+# Add any paths that contain custom static files (such as style sheets) here,
+# relative to this directory. They are copied after the builtin static files,
+# so a file named "default.css" will overwrite the builtin "default.css".
+html_static_path = ['_static']
+
+# -- Extension configuration -------------------------------------------------
+
+autodoc_inherit_docstrings = False
+napoleon_google_docstring = True
+napoleon_include_init_with_doc = True
+napoleon_numpy_docstring = False
+
+autodoc_mock_imports = [
+    'torch',
+    'tqdm',
+    'numpy',
+    'timm',
+    'pretrainedmodels',
+    'torchvision',
+    'efficientnet-pytorch',
+    'segmentation_models_pytorch.encoders',
+    'segmentation_models_pytorch.utils',
+    # 'segmentation_models_pytorch.base',
+]
+
+autoclass_content = 'both'
+autodoc_typehints = 'description'
+
+# --- Work around to make autoclass signatures not (*args, **kwargs) ----------
+
+class FakeSignature():
+    def __getattribute__(self, *args):
+        raise ValueError
+
+def f(app, obj, bound_method):
+    if "__new__" in obj.__name__:
+        obj.__signature__ = FakeSignature()
+
+def setup(app):
+    app.connect('autodoc-before-process-signature', f)

segmentation_models_pytorch/docs/encoders.rst

@@ -0,0 +1,301 @@
+🏔 Available Encoders
+=====================
+
+ResNet
+~~~~~~
+
++-------------+-------------------------+-------------+
+| Encoder     | Weights                 | Params, M   |
++=============+=========================+=============+
+| resnet18    | imagenet / ssl / swsl   | 11M         |
++-------------+-------------------------+-------------+
+| resnet34    | imagenet                | 21M         |
++-------------+-------------------------+-------------+
+| resnet50    | imagenet / ssl / swsl   | 23M         |
++-------------+-------------------------+-------------+
+| resnet101   | imagenet                | 42M         |
++-------------+-------------------------+-------------+
+| resnet152   | imagenet                | 58M         |
++-------------+-------------------------+-------------+
+
+ResNeXt
+~~~~~~~
+
++----------------------+-------------------------------------+-------------+
+| Encoder              | Weights                             | Params, M   |
++======================+=====================================+=============+
+| resnext50\_32x4d     | imagenet / ssl / swsl               | 22M         |
++----------------------+-------------------------------------+-------------+
+| resnext101\_32x4d    | ssl / swsl                          | 42M         |
++----------------------+-------------------------------------+-------------+
+| resnext101\_32x8d    | imagenet / instagram / ssl / swsl   | 86M         |
++----------------------+-------------------------------------+-------------+
+| resnext101\_32x16d   | instagram / ssl / swsl              | 191M        |
++----------------------+-------------------------------------+-------------+
+| resnext101\_32x32d   | instagram                           | 466M        |
++----------------------+-------------------------------------+-------------+
+| resnext101\_32x48d   | instagram                           | 826M        |
++----------------------+-------------------------------------+-------------+
+
+ResNeSt
+~~~~~~~
+
++----------------------------+------------+-------------+
+| Encoder                    | Weights    | Params, M   |
++============================+============+=============+
+| timm-resnest14d            | imagenet   | 8M          |
++----------------------------+------------+-------------+
+| timm-resnest26d            | imagenet   | 15M         |
++----------------------------+------------+-------------+
+| timm-resnest50d            | imagenet   | 25M         |
++----------------------------+------------+-------------+
+| timm-resnest101e           | imagenet   | 46M         |
++----------------------------+------------+-------------+
+| timm-resnest200e           | imagenet   | 68M         |
++----------------------------+------------+-------------+
+| timm-resnest269e           | imagenet   | 108M        |
++----------------------------+------------+-------------+
+| timm-resnest50d\_4s2x40d   | imagenet   | 28M         |
++----------------------------+------------+-------------+
+| timm-resnest50d\_1s4x24d   | imagenet   | 23M         |
++----------------------------+------------+-------------+
+
+Res2Ne(X)t
+~~~~~~~~~~
+
++----------------------------+------------+-------------+
+| Encoder                    | Weights    | Params, M   |
++============================+============+=============+
+| timm-res2net50\_26w\_4s    | imagenet   | 23M         |
++----------------------------+------------+-------------+
+| timm-res2net101\_26w\_4s   | imagenet   | 43M         |
++----------------------------+------------+-------------+
+| timm-res2net50\_26w\_6s    | imagenet   | 35M         |
++----------------------------+------------+-------------+
+| timm-res2net50\_26w\_8s    | imagenet   | 46M         |
++----------------------------+------------+-------------+
+| timm-res2net50\_48w\_2s    | imagenet   | 23M         |
++----------------------------+------------+-------------+
+| timm-res2net50\_14w\_8s    | imagenet   | 23M         |
++----------------------------+------------+-------------+
+| timm-res2next50            | imagenet   | 22M         |
++----------------------------+------------+-------------+
+
+RegNet(x/y)
+~~~~~~~~~~~
+
++---------------------+------------+-------------+
+| Encoder             | Weights    | Params, M   |
++=====================+============+=============+
+| timm-regnetx\_002   | imagenet   | 2M          |
++---------------------+------------+-------------+
+| timm-regnetx\_004   | imagenet   | 4M          |
++---------------------+------------+-------------+
+| timm-regnetx\_006   | imagenet   | 5M          |
++---------------------+------------+-------------+
+| timm-regnetx\_008   | imagenet   | 6M          |
++---------------------+------------+-------------+
+| timm-regnetx\_016   | imagenet   | 8M          |
++---------------------+------------+-------------+
+| timm-regnetx\_032   | imagenet   | 14M         |
++---------------------+------------+-------------+
+| timm-regnetx\_040   | imagenet   | 20M         |
++---------------------+------------+-------------+
+| timm-regnetx\_064   | imagenet   | 24M         |
++---------------------+------------+-------------+
+| timm-regnetx\_080   | imagenet   | 37M         |
++---------------------+------------+-------------+
+| timm-regnetx\_120   | imagenet   | 43M         |
++---------------------+------------+-------------+
+| timm-regnetx\_160   | imagenet   | 52M         |
++---------------------+------------+-------------+
+| timm-regnetx\_320   | imagenet   | 105M        |
++---------------------+------------+-------------+
+| timm-regnety\_002   | imagenet   | 2M          |
++---------------------+------------+-------------+
+| timm-regnety\_004   | imagenet   | 3M          |
++---------------------+------------+-------------+
+| timm-regnety\_006   | imagenet   | 5M          |
++---------------------+------------+-------------+
+| timm-regnety\_008   | imagenet   | 5M          |
++---------------------+------------+-------------+
+| timm-regnety\_016   | imagenet   | 10M         |
++---------------------+------------+-------------+
+| timm-regnety\_032   | imagenet   | 17M         |
++---------------------+------------+-------------+
+| timm-regnety\_040   | imagenet   | 19M         |
++---------------------+------------+-------------+
+| timm-regnety\_064   | imagenet   | 29M         |
++---------------------+------------+-------------+
+| timm-regnety\_080   | imagenet   | 37M         |
++---------------------+------------+-------------+
+| timm-regnety\_120   | imagenet   | 49M         |
++---------------------+------------+-------------+
+| timm-regnety\_160   | imagenet   | 80M         |
++---------------------+------------+-------------+
+| timm-regnety\_320   | imagenet   | 141M        |
++---------------------+------------+-------------+
+
+SE-Net
+~~~~~~
+
++-------------------------+------------+-------------+
+| Encoder                 | Weights    | Params, M   |
++=========================+============+=============+
+| senet154                | imagenet   | 113M        |
++-------------------------+------------+-------------+
+| se\_resnet50            | imagenet   | 26M         |
++-------------------------+------------+-------------+
+| se\_resnet101           | imagenet   | 47M         |
++-------------------------+------------+-------------+
+| se\_resnet152           | imagenet   | 64M         |
++-------------------------+------------+-------------+
+| se\_resnext50\_32x4d    | imagenet   | 25M         |
++-------------------------+------------+-------------+
+| se\_resnext101\_32x4d   | imagenet   | 46M         |
++-------------------------+------------+-------------+
+
+SK-ResNe(X)t
+~~~~~~~~~~~~
+
++---------------------------+------------+-------------+
+| Encoder                   | Weights    | Params, M   |
++===========================+============+=============+
+| timm-skresnet18           | imagenet   | 11M         |
++---------------------------+------------+-------------+
+| timm-skresnet34           | imagenet   | 21M         |
++---------------------------+------------+-------------+
+| timm-skresnext50\_32x4d   | imagenet   | 25M         |
++---------------------------+------------+-------------+
+
+DenseNet
+~~~~~~~~
+
++---------------+------------+-------------+
+| Encoder       | Weights    | Params, M   |
++===============+============+=============+
+| densenet121   | imagenet   | 6M          |
++---------------+------------+-------------+
+| densenet169   | imagenet   | 12M         |
++---------------+------------+-------------+
+| densenet201   | imagenet   | 18M         |
++---------------+------------+-------------+
+| densenet161   | imagenet   | 26M         |
++---------------+------------+-------------+
+
+Inception
+~~~~~~~~~
+
++---------------------+----------------------------------+-------------+
+| Encoder             | Weights                          | Params, M   |
++=====================+==================================+=============+
+| inceptionresnetv2   | imagenet / imagenet+background   | 54M         |
++---------------------+----------------------------------+-------------+
+| inceptionv4         | imagenet / imagenet+background   | 41M         |
++---------------------+----------------------------------+-------------+
+| xception            | imagenet                         | 22M         |
++---------------------+----------------------------------+-------------+
+
+EfficientNet
+~~~~~~~~~~~~
+
++------------------------+--------------------------------------+-------------+
+| Encoder                | Weights                              | Params, M   |
++========================+======================================+=============+
+| efficientnet-b0        | imagenet                             | 4M          |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b1        | imagenet                             | 6M          |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b2        | imagenet                             | 7M          |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b3        | imagenet                             | 10M         |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b4        | imagenet                             | 17M         |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b5        | imagenet                             | 28M         |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b6        | imagenet                             | 40M         |
++------------------------+--------------------------------------+-------------+
+| efficientnet-b7        | imagenet                             | 63M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b0   | imagenet / advprop / noisy-student   | 4M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b1   | imagenet / advprop / noisy-student   | 6M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b2   | imagenet / advprop / noisy-student   | 7M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b3   | imagenet / advprop / noisy-student   | 10M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b4   | imagenet / advprop / noisy-student   | 17M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b5   | imagenet / advprop / noisy-student   | 28M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b6   | imagenet / advprop / noisy-student   | 40M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b7   | imagenet / advprop / noisy-student   | 63M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-b8   | imagenet / advprop                   | 84M         |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-l2   | noisy-student                        | 474M        |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-lite0| imagenet                             | 4M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-lite1| imagenet                             | 4M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-lite2| imagenet                             | 6M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-lite3| imagenet                             | 8M          |
++------------------------+--------------------------------------+-------------+
+| timm-efficientnet-lite4| imagenet                             | 13M         |
++------------------------+--------------------------------------+-------------+
+
+MobileNet
+~~~~~~~~~
+
++-----------------+------------+-------------+
+| Encoder         | Weights    | Params, M   |
++=================+============+=============+
+| mobilenet\_v2   | imagenet   | 2M          |
++-----------------+------------+-------------+
+
+DPN
+~~~
+
++-----------+---------------+-------------+
+| Encoder   | Weights       | Params, M   |
++===========+===============+=============+
+| dpn68     | imagenet      | 11M         |
++-----------+---------------+-------------+
+| dpn68b    | imagenet+5k   | 11M         |
++-----------+---------------+-------------+
+| dpn92     | imagenet+5k   | 34M         |
++-----------+---------------+-------------+
+| dpn98     | imagenet      | 58M         |
++-----------+---------------+-------------+
+| dpn107    | imagenet+5k   | 84M         |
++-----------+---------------+-------------+
+| dpn131    | imagenet      | 76M         |
++-----------+---------------+-------------+
+
+VGG
+~~~
+
++-------------+------------+-------------+
+| Encoder     | Weights    | Params, M   |
++=============+============+=============+
+| vgg11       | imagenet   | 9M          |
++-------------+------------+-------------+
+| vgg11\_bn   | imagenet   | 9M          |
++-------------+------------+-------------+
+| vgg13       | imagenet   | 9M          |
++-------------+------------+-------------+
+| vgg13\_bn   | imagenet   | 9M          |
++-------------+------------+-------------+
+| vgg16       | imagenet   | 14M         |
++-------------+------------+-------------+
+| vgg16\_bn   | imagenet   | 14M         |
++-------------+------------+-------------+
+| vgg19       | imagenet   | 20M         |
++-------------+------------+-------------+
+| vgg19\_bn   | imagenet   | 20M         |
++-------------+------------+-------------+

segmentation_models_pytorch/docs/index.rst

@@ -0,0 +1,26 @@
+.. Segmentation Models documentation master file, created by
+   sphinx-quickstart on Fri Nov 27 00:00:20 2020.
+   You can adapt this file completely to your liking, but it should at least
+   contain the root `toctree` directive.
+
+Welcome to Segmentation Models's documentation!
+===============================================
+
+.. toctree::
+   :maxdepth: 2
+   :caption: Contents:
+
+   install
+   quickstart
+   models
+   encoders
+   losses
+   insights
+
+
+Indices and tables
+==================
+
+* :ref:`genindex`
+* :ref:`modindex`
+* :ref:`search`

segmentation_models_pytorch/docs/insights.rst

@@ -0,0 +1,119 @@
+🔧 Insights
+===========
+
+1. Models architecture
+~~~~~~~~~~~~~~~~~~~~~~
+
+All segmentation models in SMP (this library short name) are made of:
+
+- encoder (feature extractor, a.k.a backbone)
+- decoder (features fusion block to create segmentation *mask*)
+- segmentation head (final head to reduce number of channels from decoder and upsample mask to preserve input-output spatial resolution identity)
+- classification head (optional head which build on top of deepest encoder features)
+
+
+2. Creating your own encoder
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Encoder is a "classification model" which extract features from image and pass it to decoder.
+Each encoder should have following attributes and methods and be inherited from `segmentation_models_pytorch.encoders._base.EncoderMixin`
+
+.. code-block:: python
+
+    class MyEncoder(torch.nn.Module, EncoderMixin):
+        
+        def __init__(self, **kwargs):
+            super().__init__()
+            
+            # A number of channels for each encoder feature tensor, list of integers
+            self._out_channels: List[int] = [3, 16, 64, 128, 256, 512]
+
+            # A number of stages in decoder (in other words number of downsampling operations), integer
+            # use in in forward pass to reduce number of returning features
+            self._depth: int = 5 
+
+            # Default number of input channels in first Conv2d layer for encoder (usually 3)
+            self._in_channels: int = 3 
+            
+            # Define encoder modules below
+            ...
+
+        def forward(self, x: torch.Tensor) -> List[torch.Tensor]:
+            """Produce list of features of different spatial resolutions, each feature is a 4D torch.tensor of
+            shape NCHW (features should be sorted in descending order according to spatial resolution, starting
+            with resolution same as input `x` tensor).
+
+            Input: `x` with shape (1, 3, 64, 64)
+            Output: [f0, f1, f2, f3, f4, f5] - features with corresponding shapes
+                    [(1, 3, 64, 64), (1, 64, 32, 32), (1, 128, 16, 16), (1, 256, 8, 8),
+                    (1, 512, 4, 4), (1, 1024, 2, 2)] (C - dim may differ)
+
+            also should support number of features according to specified depth, e.g. if depth = 5,
+            number of feature tensors = 6 (one with same resolution as input and 5 downsampled),
+            depth = 3 -> number of feature tensors = 4 (one with same resolution as input and 3 downsampled).
+            """
+
+            return [feat1, feat2, feat3, feat4, feat5, feat6]
+
+When you write your own Encoder class register its build parameters
+
+.. code-block:: python
+
+    smp.encoders.encoders["my_awesome_encoder"] = {
+        "encoder": MyEncoder, # encoder class here
+        "pretrained_settings": {
+            "imagenet": {
+                "mean": [0.485, 0.456, 0.406],
+                "std": [0.229, 0.224, 0.225],
+                "url": "https://some-url.com/my-model-weights",
+                "input_space": "RGB",
+                "input_range": [0, 1],
+            },
+        },
+        "params": {
+            # init params for encoder if any
+        },
+    },
+
+Now you can use your encoder
+
+.. code-block:: python
+
+    model = smp.Unet(encoder_name="my_awesome_encoder")
+
+For better understanding see more examples of encoder in smp.encoders module.
+
+.. note::
+
+    If it works fine, don`t forget to contribute your work and make a PR to SMP 😉
+
+3. Aux classification output
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+All models support ``aux_params`` parameter, which is default set to ``None``. 
+If ``aux_params = None`` than classification auxiliary output is not created, else
+model produce not only ``mask``, but also ``label`` output with shape ``(N, C)``.
+
+Classification head consist of following layers:
+    
+1. GlobalPooling
+2. Dropout (optional)
+3. Linear
+4. Activation (optional)
+
+Example:
+
+.. code-block:: python
+    
+    aux_params=dict(
+        pooling='avg',             # one of 'avg', 'max'
+        dropout=0.5,               # dropout ratio, default is None
+        activation='sigmoid',      # activation function, default is None
+        classes=4,                 # define number of output labels
+    )
+
+    model = smp.Unet('resnet34', classes=4, aux_params=aux_params)
+    mask, label = model(x)
+
+    mask.shape, label.shape
+    # (N, 4, H, W), (N, 4)

segmentation_models_pytorch/docs/install.rst

@@ -0,0 +1,8 @@
+🛠 Installation
+===============
+
+Latest version from source:
+
+.. code-block:: bash
+
+    $ pip install -U git+https://github.com/jlcsilva/segmentation_models.pytorch

segmentation_models_pytorch/docs/losses.rst

@@ -0,0 +1,34 @@
+📉 Losses
+=========
+
+Collection of popular semantic segmentation losses. Adapted from 
+an awesome repo with pytorch utils https://github.com/BloodAxe/pytorch-toolbelt
+
+Constants
+~~~~~~~~~
+.. automodule:: segmentation_models_pytorch.losses.constants
+        :members:
+
+JaccardLoss
+~~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.JaccardLoss
+
+DiceLoss
+~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.DiceLoss
+
+FocalLoss
+~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.FocalLoss
+
+LovaszLoss
+~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.LovaszLoss
+
+SoftBCEWithLogitsLoss
+~~~~~~~~~~~~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.SoftBCEWithLogitsLoss
+
+SoftCrossEntropyLoss
+~~~~~~~~~~~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.losses.SoftCrossEntropyLoss

segmentation_models_pytorch/docs/make.bat

@@ -0,0 +1,35 @@
+@ECHO OFF
+
+pushd %~dp0
+
+REM Command file for Sphinx documentation
+
+if "%SPHINXBUILD%" == "" (
+	set SPHINXBUILD=sphinx-build
+)
+set SOURCEDIR=source
+set BUILDDIR=build
+
+if "%1" == "" goto help
+
+%SPHINXBUILD% >NUL 2>NUL
+if errorlevel 9009 (
+	echo.
+	echo.The 'sphinx-build' command was not found. Make sure you have Sphinx
+	echo.installed, then set the SPHINXBUILD environment variable to point
+	echo.to the full path of the 'sphinx-build' executable. Alternatively you
+	echo.may add the Sphinx directory to PATH.
+	echo.
+	echo.If you don't have Sphinx installed, grab it from
+	echo.http://sphinx-doc.org/
+	exit /b 1
+)
+
+%SPHINXBUILD% -M %1 %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+goto end
+
+:help
+%SPHINXBUILD% -M help %SOURCEDIR% %BUILDDIR% %SPHINXOPTS% %O%
+
+:end
+popd

segmentation_models_pytorch/docs/models.rst

@@ -0,0 +1,52 @@
+📦 Segmentation Models
+==============================
+
+Unet
+~~~~
+.. autoclass:: segmentation_models_pytorch.Unet
+
+Unet++
+~~~~~~
+.. autoclass:: segmentation_models_pytorch.UnetPlusPlus
+
+EfficientUNet++
+~~~~~~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.EfficientUnetPlusPlus
+
+ResUnet
+~~~~~~~
+.. autoclass:: segmentation_models_pytorch.ResUnet
+
+ResUnet++
+~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.ResUnetPlusPlus
+
+MAnet
+~~~~~~
+.. autoclass:: segmentation_models_pytorch.MAnet
+
+Linknet
+~~~~~~~
+.. autoclass:: segmentation_models_pytorch.Linknet
+
+FPN
+~~~
+.. autoclass:: segmentation_models_pytorch.FPN
+
+PSPNet
+~~~~~~
+.. autoclass:: segmentation_models_pytorch.PSPNet
+
+PAN
+~~~
+.. autoclass:: segmentation_models_pytorch.PAN
+
+DeepLabV3
+~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.DeepLabV3
+
+DeepLabV3+
+~~~~~~~~~~
+.. autoclass:: segmentation_models_pytorch.DeepLabV3Plus
+
+

segmentation_models_pytorch/docs/quickstart.rst

@@ -0,0 +1,36 @@
+⏳ Quick Start
+==============
+
+**1. Create segmentation model**
+
+Segmentation model is just a PyTorch nn.Module, which can be created as easy as:
+
+.. code-block:: python
+    
+    import segmentation_models_pytorch as smp
+
+    model = smp.Unet(
+        encoder_name="resnet34",        # choose encoder, e.g. mobilenet_v2 or efficientnet-b7
+        encoder_weights="imagenet",     # use `imagenet` pre-trained weights for encoder initialization
+        in_channels=1,                  # model input channels (1 for gray-scale images, 3 for RGB, etc.)
+        classes=3,                      # model output channels (number of classes in your dataset)
+    )
+
+- see table with available model architectures
+- see table with avaliable encoders and its corresponding weights
+
+**2. Configure data preprocessing**
+
+All encoders have pretrained weights. Preparing your data the same way as during weights pre-training may give your better results (higher metric score and faster convergence). But it is relevant only for 1-2-3-channels images and **not necessary** in case you train the whole model, not only decoder.
+
+.. code-block:: python
+
+    from segmentation_models_pytorch.encoders import get_preprocessing_fn
+
+    preprocess_input = get_preprocessing_fn('resnet18', pretrained='imagenet')
+
+
+**3. Congratulations!** 🎉
+
+
+You are done! Now you can train your model with your favorite framework!

segmentation_models_pytorch/docs/requirements.txt

@@ -0,0 +1,2 @@
+faculty-sphinx-theme==0.2.2
+six==1.15.0

segmentation_models_pytorch/misc/generate_table.py

@@ -0,0 +1,33 @@
+import segmentation_models_pytorch as smp
+
+encoders = smp.encoders.encoders
+
+
+WIDTH = 32
+COLUMNS = [
+    "Encoder",
+    "Weights",
+    "Params, M",
+]
+
+def wrap_row(r):
+    return "|{}|".format(r)
+
+header = "|".join([column.ljust(WIDTH, ' ') for column in COLUMNS])
+separator = "|".join(["-" * WIDTH] + [":" + "-" * (WIDTH - 2) + ":"] * (len(COLUMNS) - 1))
+
+print(wrap_row(header))
+print(wrap_row(separator))
+
+for encoder_name, encoder in encoders.items():
+    weights = "<br>".join(encoder["pretrained_settings"].keys())
+    encoder_name = encoder_name.ljust(WIDTH, " ")
+    weights = weights.ljust(WIDTH, " ")
+
+    model = encoder["encoder"](**encoder["params"], depth=5)
+    params = sum(p.numel() for p in model.parameters())
+    params = str(params // 1000000) + "M"
+    params = params.ljust(WIDTH, " ")
+
+    row = "|".join([encoder_name, weights, params])
+    print(wrap_row(row))

segmentation_models_pytorch/requirements.txt

@@ -0,0 +1,4 @@
+torchvision>=0.3.0
+pretrainedmodels==0.7.4
+efficientnet-pytorch==0.6.3
+timm==0.3.2

segmentation_models_pytorch/segmentation_models_pytorch/__init__.py

@@ -0,0 +1,49 @@
+from .unet import Unet
+from .unetplusplus import UnetPlusPlus
+from .manet import MAnet
+from .linknet import Linknet
+from .fpn import FPN
+from .pspnet import PSPNet
+from .deeplabv3 import DeepLabV3, DeepLabV3Plus
+from .pan import PAN
+from .resunet import ResUnet
+from .resunetplusplus import ResUnetPlusPlus
+from .efficientunetplusplus import EfficientUnetPlusPlus
+
+from . import encoders
+from . import utils
+from . import losses
+
+from .__version__ import __version__
+
+from typing import Optional
+import torch
+
+
+def create_model(
+    arch: str,
+    encoder_name: str = "resnet34",
+    encoder_weights: Optional[str] = "imagenet",
+    in_channels: int = 3,
+    classes: int = 1,
+    **kwargs,
+) -> torch.nn.Module:
+    """Models wrapper. Allows to create any model just with parametes
+
+    """
+
+    archs = [Unet, UnetPlusPlus, MAnet, Linknet, FPN, PSPNet, DeepLabV3, DeepLabV3Plus, PAN, ResUnet, EfficientUnetPlusPlus, ResUnetPlusPlus]
+    archs_dict = {a.__name__.lower(): a for a in archs}
+    try:
+        model_class = archs_dict[arch.lower()]
+    except KeyError:
+        raise KeyError("Wrong architecture type `{}`. Avalibale options are: {}".format(
+            arch, list(archs_dict.keys()),
+        ))
+    return model_class(
+        encoder_name=encoder_name,
+        encoder_weights=encoder_weights,
+        in_channels=in_channels,
+        classes=classes,
+        **kwargs,
+    )

segmentation_models_pytorch/segmentation_models_pytorch/__version__.py

@@ -0,0 +1,3 @@
+VERSION = (0, 1, 3)
+
+__version__ = '.'.join(map(str, VERSION))

segmentation_models_pytorch/segmentation_models_pytorch/base/__init__.py

@@ -0,0 +1,12 @@
+from .model import SegmentationModel
+
+from .modules import (
+    PreActivatedConv2dReLU,
+    Conv2dReLU,
+    Attention,
+)
+
+from .heads import (
+    SegmentationHead,
+    ClassificationHead,
+)

segmentation_models_pytorch/segmentation_models_pytorch/base/heads.py

@@ -0,0 +1,24 @@
+import torch.nn as nn
+from .modules import Flatten, Activation
+
+
+class SegmentationHead(nn.Sequential):
+
+    def __init__(self, in_channels, out_channels, kernel_size=3, activation=None, upsampling=1):
+        conv2d = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2)
+        upsampling = nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity()
+        activation = Activation(activation)
+        super().__init__(conv2d, upsampling, activation)
+
+
+class ClassificationHead(nn.Sequential):
+
+    def __init__(self, in_channels, classes, pooling="avg", dropout=0.2, activation=None):
+        if pooling not in ("max", "avg"):
+            raise ValueError("Pooling should be one of ('max', 'avg'), got {}.".format(pooling))
+        pool = nn.AdaptiveAvgPool2d(1) if pooling == 'avg' else nn.AdaptiveMaxPool2d(1)
+        flatten = Flatten()
+        dropout = nn.Dropout(p=dropout, inplace=True) if dropout else nn.Identity()
+        linear = nn.Linear(in_channels, classes, bias=True)
+        activation = Activation(activation)
+        super().__init__(pool, flatten, dropout, linear, activation)

segmentation_models_pytorch/segmentation_models_pytorch/base/initialization.py

@@ -0,0 +1,27 @@
+import torch.nn as nn
+
+
+def initialize_decoder(module):
+    for m in module.modules():
+
+        if isinstance(m, nn.Conv2d):
+            nn.init.kaiming_uniform_(m.weight, mode="fan_in", nonlinearity="relu")
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+
+        elif isinstance(m, nn.BatchNorm2d):
+            nn.init.constant_(m.weight, 1)
+            nn.init.constant_(m.bias, 0)
+
+        elif isinstance(m, nn.Linear):
+            nn.init.xavier_uniform_(m.weight)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+
+
+def initialize_head(module):
+    for m in module.modules():
+        if isinstance(m, (nn.Linear, nn.Conv2d)):
+            nn.init.xavier_uniform_(m.weight)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)

segmentation_models_pytorch/segmentation_models_pytorch/base/model.py

@@ -0,0 +1,42 @@
+import torch
+from . import initialization as init
+
+
+class SegmentationModel(torch.nn.Module):
+
+    def initialize(self):
+        init.initialize_decoder(self.decoder)
+        init.initialize_head(self.segmentation_head)
+        if self.classification_head is not None:
+            init.initialize_head(self.classification_head)
+
+    def forward(self, x):
+        """Sequentially pass `x` trough model`s encoder, decoder and heads"""
+        features = self.encoder(x)
+        decoder_output = self.decoder(*features)
+
+        masks = self.segmentation_head(decoder_output)
+
+        if self.classification_head is not None:
+            labels = self.classification_head(features[-1])
+            return masks, labels
+
+        return masks
+
+    def predict(self, x):
+        """Inference method. Switch model to `eval` mode, call `.forward(x)` with `torch.no_grad()`
+
+        Args:
+            x: 4D torch tensor with shape (batch_size, channels, height, width)
+
+        Return:
+            prediction: 4D torch tensor with shape (batch_size, classes, height, width)
+
+        """
+        if self.training:
+            self.eval()
+
+        with torch.no_grad():
+            x = self.forward(x)
+
+        return x

segmentation_models_pytorch/segmentation_models_pytorch/base/modules.py

@@ -0,0 +1,206 @@
+import torch
+import torch.nn as nn
+
+try:
+    from inplace_abn import InPlaceABN
+except ImportError:
+    InPlaceABN = None
+
+class PreActivatedConv2dReLU(nn.Sequential):
+    """
+    Pre-activated 2D convolution, as proposed in https://arxiv.org/pdf/1603.05027.pdf. Feature maps are processed by a normalization layer, 
+    followed by a ReLU activation and a 3x3 convolution.
+    normalization
+    """
+    def __init__(
+            self,
+            in_channels, 
+            out_channels,
+            kernel_size,
+            padding=0,
+            stride=1,
+            use_batchnorm=True,
+    ):
+
+        if use_batchnorm == "inplace" and InPlaceABN is None:
+            raise RuntimeError(
+                "In order to use `use_batchnorm='inplace'` inplace_abn package must be installed. "
+                + "To install see: https://github.com/mapillary/inplace_abn"
+            )
+        if use_batchnorm == "inplace":
+            bn = InPlaceABN(out_channels, activation="leaky_relu", activation_param=0.0)
+            relu = nn.Identity()
+        elif use_batchnorm and use_batchnorm != "inplace":
+            bn = nn.BatchNorm2d(out_channels)
+        else:
+            bn = nn.Identity()
+
+        relu = nn.ReLU(inplace=True)
+
+        conv = nn.Conv2d(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            bias=not (use_batchnorm),
+        )
+        super(PreActivatedConv2dReLU, self).__init__(conv, bn, relu)
+
+class Conv2dReLU(nn.Sequential):
+    """
+    Block composed of a 3x3 convolution followed by a normalization layer and ReLU activation.
+    """
+    def __init__(
+            self,
+            in_channels,
+            out_channels,
+            kernel_size,
+            padding=0,
+            stride=1,
+            use_batchnorm=True,
+    ):
+
+        if use_batchnorm == "inplace" and InPlaceABN is None:
+            raise RuntimeError(
+                "In order to use `use_batchnorm='inplace'` inplace_abn package must be installed. "
+                + "To install see: https://github.com/mapillary/inplace_abn"
+            )
+
+        conv = nn.Conv2d(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            bias=not (use_batchnorm),
+        )
+        relu = nn.ReLU(inplace=True)
+
+        if use_batchnorm == "inplace":
+            bn = InPlaceABN(out_channels, activation="leaky_relu", activation_param=0.0)
+            relu = nn.Identity()
+        elif use_batchnorm and use_batchnorm != "inplace":
+            bn = nn.BatchNorm2d(out_channels)
+        else:
+            bn = nn.Identity()
+
+        super(Conv2dReLU, self).__init__(conv, bn, relu)
+
+class DepthWiseConv2d(nn.Conv2d):
+    "Depth-wise convolution operation"
+    def __init__(self, channels, kernel_size=3, stride=1):
+        super().__init__(channels, channels, kernel_size, stride=stride, padding=kernel_size//2, groups=channels)
+
+class PointWiseConv2d(nn.Conv2d):
+    "Point-wise (1x1) convolution operation"
+    def __init__(self, in_channels, out_channels):
+        super().__init__(in_channels, out_channels, kernel_size=1, stride=1)
+
+class SEModule(nn.Module):
+    """
+    Spatial squeeze & channel excitation attention module, as proposed in https://arxiv.org/abs/1709.01507.
+    """
+    def __init__(self, in_channels, reduction=16):
+        super().__init__()
+        self.cSE = nn.Sequential(
+            nn.AdaptiveAvgPool2d(1),
+            nn.Conv2d(in_channels, in_channels // reduction, 1),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(in_channels // reduction, in_channels, 1),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, x):
+        return x * self.cSE(x)
+
+class sSEModule(nn.Module):
+    """
+    Channel squeeze & spatial excitation attention module, as proposed in https://arxiv.org/abs/1808.08127.
+    """
+    def __init__(self, in_channels):
+        super().__init__()
+        self.sSE = nn.Sequential(nn.Conv2d(in_channels, 1, 1), nn.Sigmoid())
+
+    def forward(self, x):
+        return x * self.sSE(x)
+
+class SCSEModule(nn.Module):
+    """
+    Concurrent spatial and channel squeeze & excitation attention module, as proposed in https://arxiv.org/pdf/1803.02579.pdf.
+    """
+    def __init__(self, in_channels, reduction=16):
+        super().__init__()
+        self.cSE = nn.Sequential(
+            nn.AdaptiveAvgPool2d(1),
+            nn.Conv2d(in_channels, in_channels // reduction, 1),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(in_channels // reduction, in_channels, 1),
+            nn.Sigmoid(),
+        )
+        self.sSE = nn.Sequential(nn.Conv2d(in_channels, 1, 1), nn.Sigmoid())
+
+    def forward(self, x):
+        return x * self.cSE(x) + x * self.sSE(x)
+
+class ArgMax(nn.Module):
+
+    def __init__(self, dim=None):
+        super().__init__()
+        self.dim = dim
+
+    def forward(self, x):
+        return torch.argmax(x, dim=self.dim)
+
+
+class Activation(nn.Module):
+
+    def __init__(self, name, **params):
+
+        super().__init__()
+
+        if name is None or name == 'identity':
+            self.activation = nn.Identity(**params)
+        elif name == 'sigmoid':
+            self.activation = nn.Sigmoid()
+        elif name == 'softmax2d':
+            self.activation = nn.Softmax(dim=1, **params)
+        elif name == 'softmax':
+            self.activation = nn.Softmax(**params)
+        elif name == 'logsoftmax':
+            self.activation = nn.LogSoftmax(**params)
+        elif name == 'tanh':
+            self.activation = nn.Tanh()
+        elif name == 'argmax':
+            self.activation = ArgMax(**params)
+        elif name == 'argmax2d':
+            self.activation = ArgMax(dim=1, **params)
+        elif callable(name):
+            self.activation = name(**params)
+        else:
+            raise ValueError('Activation should be callable/sigmoid/softmax/logsoftmax/tanh/None; got {}'.format(name))
+
+    def forward(self, x):
+        return self.activation(x)
+
+
+class Attention(nn.Module):
+
+    def __init__(self, name, **params):
+        super().__init__()
+
+        if name is None:
+            self.attention = nn.Identity(**params)
+        elif name == 'scse':
+            self.attention = SCSEModule(**params)
+        elif name == 'se':
+            self.attention = SEModule(**params)
+        else:
+            raise ValueError("Attention {} is not implemented".format(name))
+
+    def forward(self, x):
+        return self.attention(x)
+
+class Flatten(nn.Module):
+    def forward(self, x):
+        return x.view(x.shape[0], -1)

segmentation_models_pytorch/segmentation_models_pytorch/deeplabv3/__init__.py

@@ -0,0 +1 @@
+from .model import DeepLabV3, DeepLabV3Plus

segmentation_models_pytorch/segmentation_models_pytorch/deeplabv3/decoder.py

@@ -0,0 +1,220 @@
+"""
+BSD 3-Clause License
+
+Copyright (c) Soumith Chintala 2016,
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+"""
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+__all__ = ["DeepLabV3Decoder"]
+
+
+class DeepLabV3Decoder(nn.Sequential):
+    def __init__(self, in_channels, out_channels=256, atrous_rates=(12, 24, 36)):
+        super().__init__(
+            ASPP(in_channels, out_channels, atrous_rates),
+            nn.Conv2d(out_channels, out_channels, 3, padding=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+        self.out_channels = out_channels
+
+    def forward(self, *features):
+        return super().forward(features[-1])
+
+
+class DeepLabV3PlusDecoder(nn.Module):
+    def __init__(
+        self,
+        encoder_channels,
+        out_channels=256,
+        atrous_rates=(12, 24, 36),
+        output_stride=16,
+    ):
+        super().__init__()
+        if output_stride not in {8, 16}:
+            raise ValueError("Output stride should be 8 or 16, got {}.".format(output_stride))
+
+        self.out_channels = out_channels
+        self.output_stride = output_stride
+
+        self.aspp = nn.Sequential(
+            ASPP(encoder_channels[-1], out_channels, atrous_rates, separable=True),
+            SeparableConv2d(out_channels, out_channels, kernel_size=3, padding=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+
+        scale_factor = 2 if output_stride == 8 else 4
+        self.up = nn.UpsamplingBilinear2d(scale_factor=scale_factor)
+
+        highres_in_channels = encoder_channels[-4]
+        highres_out_channels = 48   # proposed by authors of paper
+        self.block1 = nn.Sequential(
+            nn.Conv2d(highres_in_channels, highres_out_channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(highres_out_channels),
+            nn.ReLU(),
+        )
+        self.block2 = nn.Sequential(
+            SeparableConv2d(
+                highres_out_channels + out_channels,
+                out_channels,
+                kernel_size=3,
+                padding=1,
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+
+    def forward(self, *features):
+        aspp_features = self.aspp(features[-1])
+        aspp_features = self.up(aspp_features)
+        high_res_features = self.block1(features[-4])
+        concat_features = torch.cat([aspp_features, high_res_features], dim=1)
+        fused_features = self.block2(concat_features)
+        return fused_features
+
+
+class ASPPConv(nn.Sequential):
+    def __init__(self, in_channels, out_channels, dilation):
+        super().__init__(
+            nn.Conv2d(
+                in_channels,
+                out_channels,
+                kernel_size=3,
+                padding=dilation,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+
+
+class ASPPSeparableConv(nn.Sequential):
+    def __init__(self, in_channels, out_channels, dilation):
+        super().__init__(
+            SeparableConv2d(
+                in_channels,
+                out_channels,
+                kernel_size=3,
+                padding=dilation,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+
+
+class ASPPPooling(nn.Sequential):
+    def __init__(self, in_channels, out_channels):
+        super().__init__(
+            nn.AdaptiveAvgPool2d(1),
+            nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+        )
+
+    def forward(self, x):
+        size = x.shape[-2:]
+        for mod in self:
+            x = mod(x)
+        return F.interpolate(x, size=size, mode='bilinear', align_corners=False)
+
+
+class ASPP(nn.Module):
+    def __init__(self, in_channels, out_channels, atrous_rates, separable=False):
+        super(ASPP, self).__init__()
+        modules = []
+        modules.append(
+            nn.Sequential(
+                nn.Conv2d(in_channels, out_channels, 1, bias=False),
+                nn.BatchNorm2d(out_channels),
+                nn.ReLU(),
+            )
+        )
+
+        rate1, rate2, rate3 = tuple(atrous_rates)
+        ASPPConvModule = ASPPConv if not separable else ASPPSeparableConv
+
+        modules.append(ASPPConvModule(in_channels, out_channels, rate1))
+        modules.append(ASPPConvModule(in_channels, out_channels, rate2))
+        modules.append(ASPPConvModule(in_channels, out_channels, rate3))
+        modules.append(ASPPPooling(in_channels, out_channels))
+
+        self.convs = nn.ModuleList(modules)
+
+        self.project = nn.Sequential(
+            nn.Conv2d(5 * out_channels, out_channels, kernel_size=1, bias=False),
+            nn.BatchNorm2d(out_channels),
+            nn.ReLU(),
+            nn.Dropout(0.5),
+        )
+
+    def forward(self, x):
+        res = []
+        for conv in self.convs:
+            res.append(conv(x))
+        res = torch.cat(res, dim=1)
+        return self.project(res)
+
+
+class SeparableConv2d(nn.Sequential):
+
+    def __init__(
+            self,
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride=1,
+            padding=0,
+            dilation=1,
+            bias=True,
+    ):
+        dephtwise_conv = nn.Conv2d(
+            in_channels,
+            in_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=in_channels,
+            bias=False,
+        )
+        pointwise_conv = nn.Conv2d(
+            in_channels,
+            out_channels,
+            kernel_size=1,
+            bias=bias,
+        )
+        super().__init__(dephtwise_conv, pointwise_conv)

segmentation_models_pytorch/segmentation_models_pytorch/deeplabv3/model.py

@@ -0,0 +1,183 @@
+import torch.nn as nn
+
+from typing import Optional
+from .decoder import DeepLabV3Decoder, DeepLabV3PlusDecoder
+from ..base import SegmentationModel, SegmentationHead, ClassificationHead
+from ..encoders import get_encoder
+
+
+class DeepLabV3(SegmentationModel):
+    """DeepLabV3_ implementation from "Rethinking Atrous Convolution for Semantic Image Segmentation"
+
+    Args:
+        encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
+            to extract features of different spatial resolution
+        encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
+            with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
+            Default is 5
+        encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
+            other pretrained weights (see table with available weights for each encoder_name)
+        decoder_channels: A number of convolution filters in ASPP module. Default is 256
+        in_channels: A number of input channels for the model, default is 3 (RGB images)
+        classes: A number of classes for output mask (or you can think as a number of channels of output mask)
+        activation: An activation function to apply after the final convolution layer.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
+            Default is **None**
+        upsampling: Final upsampling factor. Default is 8 to preserve input-output spatial shape identity
+        aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
+            on top of encoder if **aux_params** is not **None** (default). Supported params:
+                - classes (int): A number of classes
+                - pooling (str): One of "max", "avg". Default is "avg"
+                - dropout (float): Dropout factor in [0, 1)
+                - activation (str): An activation function to apply "sigmoid"/"softmax" (could be **None** to return logits)
+    Returns:
+        ``torch.nn.Module``: **DeepLabV3**
+
+    .. _DeepLabV3:
+        https://arxiv.org/abs/1706.05587
+
+    Reference:
+        https://arxiv.org/abs/1706.05587
+    """
+
+    def __init__(
+            self,
+            encoder_name: str = "resnet34",
+            encoder_depth: int = 5,
+            encoder_weights: Optional[str] = "imagenet",
+            decoder_channels: int = 256,
+            in_channels: int = 3,
+            classes: int = 1,
+            activation: Optional[str] = None,
+            upsampling: int = 8,
+            aux_params: Optional[dict] = None,
+    ):
+        super().__init__()
+
+        self.encoder = get_encoder(
+            encoder_name,
+            in_channels=in_channels,
+            depth=encoder_depth,
+            weights=encoder_weights,
+        )
+        self.encoder.make_dilated(
+            stage_list=[4, 5],
+            dilation_list=[2, 4]
+        )
+
+        self.decoder = DeepLabV3Decoder(
+            in_channels=self.encoder.out_channels[-1],
+            out_channels=decoder_channels,
+        )
+
+        self.segmentation_head = SegmentationHead(
+            in_channels=self.decoder.out_channels,
+            out_channels=classes,
+            activation=activation,
+            kernel_size=1,
+            upsampling=upsampling,
+        )
+
+        if aux_params is not None:
+            self.classification_head = ClassificationHead(
+                in_channels=self.encoder.out_channels[-1], **aux_params
+            )
+        else:
+            self.classification_head = None
+
+
+class DeepLabV3Plus(SegmentationModel):
+    """DeepLabV3+ implementation from "Encoder-Decoder with Atrous Separable
+    Convolution for Semantic Image Segmentation"
+    
+    Args:
+        encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
+            to extract features of different spatial resolution
+        encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
+            with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
+            Default is 5
+        encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
+            other pretrained weights (see table with available weights for each encoder_name)
+        encoder_output_stride: Downsampling factor for last encoder features (see original paper for explanation)
+        decoder_atrous_rates: Dilation rates for ASPP module (should be a tuple of 3 integer values)
+        decoder_channels: A number of convolution filters in ASPP module. Default is 256
+        in_channels: A number of input channels for the model, default is 3 (RGB images)
+        classes: A number of classes for output mask (or you can think as a number of channels of output mask)
+        activation: An activation function to apply after the final convolution layer.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
+            Default is **None**
+        upsampling: Final upsampling factor. Default is 4 to preserve input-output spatial shape identity
+        aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
+            on top of encoder if **aux_params** is not **None** (default). Supported params:
+                - classes (int): A number of classes
+                - pooling (str): One of "max", "avg". Default is "avg"
+                - dropout (float): Dropout factor in [0, 1)
+                - activation (str): An activation function to apply "sigmoid"/"softmax" (could be **None** to return logits)
+    Returns:
+        ``torch.nn.Module``: **DeepLabV3Plus**
+
+    Reference:
+        https://arxiv.org/abs/1802.02611v3
+    """
+    def __init__(
+            self,
+            encoder_name: str = "resnet34",
+            encoder_depth: int = 5,
+            encoder_weights: Optional[str] = "imagenet",
+            encoder_output_stride: int = 16,
+            decoder_channels: int = 256,
+            decoder_atrous_rates: tuple = (12, 24, 36),
+            in_channels: int = 3,
+            classes: int = 1,
+            activation: Optional[str] = None,
+            upsampling: int = 4,
+            aux_params: Optional[dict] = None,
+    ):
+        super().__init__()
+
+        self.encoder = get_encoder(
+            encoder_name,
+            in_channels=in_channels,
+            depth=encoder_depth,
+            weights=encoder_weights,
+        )
+
+        if encoder_output_stride == 8:
+            self.encoder.make_dilated(
+                stage_list=[4, 5],
+                dilation_list=[2, 4]
+            )
+
+        elif encoder_output_stride == 16:
+            self.encoder.make_dilated(
+                stage_list=[5],
+                dilation_list=[2]
+            )
+        else:
+            raise ValueError(
+                "Encoder output stride should be 8 or 16, got {}".format(encoder_output_stride)
+            )
+
+        self.decoder = DeepLabV3PlusDecoder(
+            encoder_channels=self.encoder.out_channels,
+            out_channels=decoder_channels,
+            atrous_rates=decoder_atrous_rates,
+            output_stride=encoder_output_stride,
+        )
+
+        self.segmentation_head = SegmentationHead(
+            in_channels=self.decoder.out_channels,
+            out_channels=classes,
+            activation=activation,
+            kernel_size=1,
+            upsampling=upsampling,
+        )
+
+        if aux_params is not None:
+            self.classification_head = ClassificationHead(
+                in_channels=self.encoder.out_channels[-1], **aux_params
+            )
+        else:
+            self.classification_head = None

segmentation_models_pytorch/segmentation_models_pytorch/efficientunetplusplus/__init__.py

@@ -0,0 +1 @@
+from .model import EfficientUnetPlusPlus

segmentation_models_pytorch/segmentation_models_pytorch/efficientunetplusplus/decoder.py

@@ -0,0 +1,148 @@
+import torch
+from torch.functional import norm
+import torch.nn as nn
+import torch.nn.functional as F
+
+from ..base import modules as md
+
+class InvertedResidual(nn.Module):
+    """
+    Inverted bottleneck residual block with an scSE block embedded into the residual layer, after the 
+    depthwise convolution. By default, uses batch normalization and Hardswish activation.
+    """
+    def __init__(self, in_channels, out_channels, kernel_size = 3, stride = 1, expansion_ratio = 1, squeeze_ratio = 1, \
+        activation = nn.Hardswish(True), normalization = nn.BatchNorm2d):
+        super().__init__()
+        self.same_shape = in_channels == out_channels
+        self.mid_channels = expansion_ratio*in_channels
+        self.block = nn.Sequential(
+            md.PointWiseConv2d(in_channels, self.mid_channels),
+            normalization(self.mid_channels),
+            activation,
+            md.DepthWiseConv2d(self.mid_channels, kernel_size=kernel_size, stride=stride),
+            normalization(self.mid_channels),
+            activation,
+            #md.sSEModule(self.mid_channels),
+            md.SCSEModule(self.mid_channels, reduction = squeeze_ratio),
+            #md.SEModule(self.mid_channels, reduction = squeeze_ratio),
+            md.PointWiseConv2d(self.mid_channels, out_channels),
+            normalization(out_channels)
+        )
+        
+        if not self.same_shape:
+            # 1x1 convolution used to match the number of channels in the skip feature maps with that 
+            # of the residual feature maps
+            self.skip_conv = nn.Sequential(
+                nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=1),
+                normalization(out_channels)
+            )
+          
+    def forward(self, x):
+        residual = self.block(x)
+        
+        if not self.same_shape:
+            x = self.skip_conv(x)
+        return x + residual
+        
+class DecoderBlock(nn.Module):
+    def __init__(
+            self,
+            in_channels,
+            skip_channels,
+            out_channels,
+            squeeze_ratio=1,
+            expansion_ratio=1
+    ):
+        super().__init__()
+
+        # Inverted Residual block convolutions
+        self.conv1 = InvertedResidual(
+            in_channels=in_channels+skip_channels, 
+            out_channels=out_channels, 
+            kernel_size=3, 
+            stride=1, 
+            expansion_ratio=expansion_ratio, 
+            squeeze_ratio=squeeze_ratio
+        )
+        self.conv2 = InvertedResidual(
+            in_channels=out_channels, 
+            out_channels=out_channels, 
+            kernel_size=3, 
+            stride=1, 
+            expansion_ratio=expansion_ratio, 
+            squeeze_ratio=squeeze_ratio
+        )
+
+    def forward(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="nearest")
+
+        if skip is not None:
+            x = torch.cat([x, skip], dim=1)
+        x = self.conv1(x)
+        x = self.conv2(x)
+        return x
+
+class EfficientUnetPlusPlusDecoder(nn.Module):
+    def __init__(
+            self,
+            encoder_channels,
+            decoder_channels,
+            n_blocks=5,
+            squeeze_ratio=1,
+            expansion_ratio=1
+    ):
+        super().__init__()
+        if n_blocks != len(decoder_channels):
+            raise ValueError(
+                "Model depth is {}, but you provide `decoder_channels` for {} blocks.".format(
+                    n_blocks, len(decoder_channels)
+                )
+            )
+
+        encoder_channels = encoder_channels[1:]  # remove first skip with same spatial resolution
+        encoder_channels = encoder_channels[::-1]  # reverse channels to start from head of encoder
+        # computing blocks input and output channels
+        head_channels = encoder_channels[0]
+        self.in_channels = [head_channels] + list(decoder_channels[:-1])
+        self.skip_channels = list(encoder_channels[1:]) + [0]
+        self.out_channels = decoder_channels
+
+        # combine decoder keyword arguments
+        kwargs = dict(squeeze_ratio=squeeze_ratio, expansion_ratio=expansion_ratio)
+
+        blocks = {}
+        for layer_idx in range(len(self.in_channels) - 1):
+            for depth_idx in range(layer_idx+1):
+                if depth_idx == 0:
+                    in_ch = self.in_channels[layer_idx]
+                    skip_ch = self.skip_channels[layer_idx] * (layer_idx+1)
+                    out_ch = self.out_channels[layer_idx]
+                else:
+                    out_ch = self.skip_channels[layer_idx]
+                    skip_ch = self.skip_channels[layer_idx] * (layer_idx+1-depth_idx)
+                    in_ch = self.skip_channels[layer_idx - 1]
+                blocks[f'x_{depth_idx}_{layer_idx}'] = DecoderBlock(in_ch, skip_ch, out_ch, **kwargs)
+        blocks[f'x_{0}_{len(self.in_channels)-1}'] =\
+            DecoderBlock(self.in_channels[-1], 0, self.out_channels[-1], **kwargs)
+        self.blocks = nn.ModuleDict(blocks)
+        self.depth = len(self.in_channels) - 1
+
+    def forward(self, *features):
+
+        features = features[1:]    # remove first skip with same spatial resolution
+        features = features[::-1]  # reverse channels to start from head of encoder
+        # start building dense connections
+        dense_x = {}
+        for layer_idx in range(len(self.in_channels)-1):
+            for depth_idx in range(self.depth-layer_idx):
+                if layer_idx == 0:
+                    output = self.blocks[f'x_{depth_idx}_{depth_idx}'](features[depth_idx], features[depth_idx+1])
+                    dense_x[f'x_{depth_idx}_{depth_idx}'] = output
+                else:
+                    dense_l_i = depth_idx + layer_idx
+                    cat_features = [dense_x[f'x_{idx}_{dense_l_i}'] for idx in range(depth_idx+1, dense_l_i+1)]
+                    cat_features = torch.cat(cat_features + [features[dense_l_i+1]], dim=1)
+                    dense_x[f'x_{depth_idx}_{dense_l_i}'] =\
+                        self.blocks[f'x_{depth_idx}_{dense_l_i}'](dense_x[f'x_{depth_idx}_{dense_l_i-1}'], cat_features)
+        dense_x[f'x_{0}_{self.depth}'] = self.blocks[f'x_{0}_{self.depth}'](dense_x[f'x_{0}_{self.depth-1}'])
+        return dense_x[f'x_{0}_{self.depth}']

segmentation_models_pytorch/segmentation_models_pytorch/efficientunetplusplus/model.py

@@ -0,0 +1,125 @@
+from typing import Optional, Union, List
+from .decoder import EfficientUnetPlusPlusDecoder
+from ..encoders import get_encoder
+from ..base import SegmentationModel
+from ..base import SegmentationHead, ClassificationHead
+from torchvision import transforms
+
+class EfficientUnetPlusPlus(SegmentationModel):
+    """The EfficientUNet++ is a fully convolutional neural network for ordinary and medical image semantic segmentation. 
+    Consists of an *encoder* and a *decoder*, connected by *skip connections*. The encoder extracts features of 
+    different spatial resolutions, which are fed to the decoder through skip connections. The decoder combines its 
+    own feature maps with the ones from skip connections to produce accurate segmentations masks.  The EfficientUNet++ 
+    decoder architecture is based on the UNet++, a model composed of nested U-Net-like decoder sub-networks. To 
+    increase performance and computational efficiency, the EfficientUNet++ replaces the UNet++'s blocks with 
+    inverted residual blocks with depthwise convolutions and embedded spatial and channel attention mechanisms.
+    Synergizes well with EfficientNet encoders. Due to their efficient visual representations (i.e., using few channels
+    to represent extracted features), EfficientNet encoders require few computation from the decoder.
+
+    Args:
+        encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone) to extract features
+        encoder_depth: Number of stages of the encoder, in range [3 ,5]. Each stage generate features two times smaller, 
+            in spatial dimensions, than the previous one (e.g., for depth=0 features will haves shapes [(N, C, H, W)]), 
+            for depth 1 features will have shapes [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
+            Default is 5
+        encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
+            other pretrained weights (see table with available weights for each encoder_name)
+        decoder_channels: List of integers which specify **in_channels** parameter for convolutions used in the decoder.
+            Length of the list should be the same as **encoder_depth**
+        in_channels: The number of input channels of the model, default is 3 (RGB images)
+        classes: The number of classes of the output mask. Can be thought of as the number of channels of the mask
+        activation: An activation function to apply after the final convolution layer.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
+            Default is **None**
+        aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is built 
+            on top of encoder if **aux_params** is not **None** (default). Supported params:
+                - classes (int): A number of classes
+                - pooling (str): One of "max", "avg". Default is "avg"
+                - dropout (float): Dropout factor in [0, 1)
+                - activation (str): An activation function to apply "sigmoid"/"softmax" (could be **None** to return logits)
+    Returns:
+        ``torch.nn.Module``: **EfficientUnet++**
+
+    Reference:
+        https://arxiv.org/abs/2106.11447
+    """
+
+    def __init__(
+        self,
+        encoder_name: str = "timm-efficientnet-b5",
+        encoder_depth: int = 5,
+        encoder_weights: Optional[str] = "imagenet",
+        decoder_channels: List[int] = (256, 128, 64, 32, 16),
+        squeeze_ratio: int = 1,
+        expansion_ratio: int = 1,
+        in_channels: int = 3,
+        classes: int = 1,
+        activation: Optional[Union[str, callable]] = None,
+        aux_params: Optional[dict] = None,
+    ):
+        super().__init__()
+        self.classes = classes
+        self.encoder = get_encoder(
+            encoder_name,
+            in_channels=in_channels,
+            depth=encoder_depth,
+            weights=encoder_weights,
+        )
+
+        self.decoder = EfficientUnetPlusPlusDecoder(
+            encoder_channels=self.encoder.out_channels,
+            decoder_channels=decoder_channels,
+            n_blocks=encoder_depth,
+            squeeze_ratio=squeeze_ratio,
+            expansion_ratio=expansion_ratio
+        )
+
+        self.segmentation_head = SegmentationHead(
+            in_channels=decoder_channels[-1],
+            out_channels=classes,
+            activation=activation,
+            kernel_size=3,
+        )
+
+        if aux_params is not None:
+            self.classification_head = ClassificationHead(
+                in_channels=self.encoder.out_channels[-1], **aux_params
+            )
+        else:
+            self.classification_head = None
+
+        self.name = "EfficientUNet++-{}".format(encoder_name)
+        self.initialize()
+
+    def predict(self, x):
+        """Inference method. Switch model to `eval` mode, call `.forward(x)` with `torch.no_grad()`
+
+        Args:
+            x: 4D torch tensor with shape (batch_size, channels, height, width)
+
+        Return:
+            prediction: 4D torch tensor with shape (batch_size, classes, height, width)
+
+        """
+        if self.training:
+            self.eval()
+
+        with torch.no_grad():
+            output = self.forward(x)
+
+        if self.classes > 1:
+            probs = torch.softmax(output, dim=1)
+        else:
+            probs = torch.sigmoid(output)
+
+        probs = probs.squeeze(0)
+        tf = transforms.Compose(
+            [
+                transforms.ToPILImage(),
+                transforms.Resize(x.size[1]),
+                transforms.ToTensor()
+            ]
+        )
+        full_mask = tf(probs.cpu())   
+
+        return full_mask

segmentation_models_pytorch/segmentation_models_pytorch/encoders/__init__.py

@@ -0,0 +1,84 @@
+import functools
+import torch.utils.model_zoo as model_zoo
+
+from .resnet import resnet_encoders
+from .dpn import dpn_encoders
+from .vgg import vgg_encoders
+from .senet import senet_encoders
+from .densenet import densenet_encoders
+from .inceptionresnetv2 import inceptionresnetv2_encoders
+from .inceptionv4 import inceptionv4_encoders
+from .efficientnet import efficient_net_encoders
+from .mobilenet import mobilenet_encoders
+from .xception import xception_encoders
+# from .timm_efficientnet import timm_efficientnet_encoders
+from .timm_resnest import timm_resnest_encoders
+from .timm_res2net import timm_res2net_encoders
+from .timm_regnet import timm_regnet_encoders
+from .timm_sknet import timm_sknet_encoders
+from ._preprocessing import preprocess_input
+
+encoders = {}
+encoders.update(resnet_encoders)
+encoders.update(dpn_encoders)
+encoders.update(vgg_encoders)
+encoders.update(senet_encoders)
+encoders.update(densenet_encoders)
+encoders.update(inceptionresnetv2_encoders)
+encoders.update(inceptionv4_encoders)
+encoders.update(efficient_net_encoders)
+encoders.update(mobilenet_encoders)
+encoders.update(xception_encoders)
+# encoders.update(timm_efficientnet_encoders)
+encoders.update(timm_resnest_encoders)
+encoders.update(timm_res2net_encoders)
+encoders.update(timm_regnet_encoders)
+encoders.update(timm_sknet_encoders)
+
+
+def get_encoder(name, in_channels=3, depth=5, weights=None):
+
+    try:
+        Encoder = encoders[name]["encoder"]
+    except KeyError:
+        raise KeyError("Wrong encoder name `{}`, supported encoders: {}".format(name, list(encoders.keys())))
+
+    params = encoders[name]["params"]
+    params.update(depth=depth)
+    encoder = Encoder(**params)
+
+    if weights is not None:
+        try:
+            settings = encoders[name]["pretrained_settings"][weights]
+        except KeyError:
+            raise KeyError("Wrong pretrained weights `{}` for encoder `{}`. Available options are: {}".format(
+                weights, name, list(encoders[name]["pretrained_settings"].keys()),
+            ))
+        encoder.load_state_dict(model_zoo.load_url(settings["url"]))
+
+    encoder.set_in_channels(in_channels)
+
+    return encoder
+
+
+def get_encoder_names():
+    return list(encoders.keys())
+
+
+def get_preprocessing_params(encoder_name, pretrained="imagenet"):
+    settings = encoders[encoder_name]["pretrained_settings"]
+
+    if pretrained not in settings.keys():
+        raise ValueError("Available pretrained options {}".format(settings.keys()))
+
+    formatted_settings = {}
+    formatted_settings["input_space"] = settings[pretrained].get("input_space")
+    formatted_settings["input_range"] = settings[pretrained].get("input_range")
+    formatted_settings["mean"] = settings[pretrained].get("mean")
+    formatted_settings["std"] = settings[pretrained].get("std")
+    return formatted_settings
+
+
+def get_preprocessing_fn(encoder_name, pretrained="imagenet"):
+    params = get_preprocessing_params(encoder_name, pretrained=pretrained)
+    return functools.partial(preprocess_input, **params)

segmentation_models_pytorch/segmentation_models_pytorch/encoders/_base.py

@@ -0,0 +1,41 @@
+import torch
+import torch.nn as nn
+from typing import List
+from collections import OrderedDict
+
+from . import _utils as utils
+
+
+class EncoderMixin:
+    """Add encoder functionality such as:
+        - output channels specification of feature tensors (produced by encoder)
+        - patching first convolution for arbitrary input channels
+    """
+
+    @property
+    def out_channels(self):
+        """Return channels dimensions for each tensor of forward output of encoder"""
+        return self._out_channels[: self._depth + 1]
+
+    def set_in_channels(self, in_channels):
+        """Change first convolution channels"""
+        if in_channels == 3:
+            return
+
+        self._in_channels = in_channels
+        if self._out_channels[0] == 3:
+            self._out_channels = tuple([in_channels] + list(self._out_channels)[1:])
+
+        utils.patch_first_conv(model=self, in_channels=in_channels)
+
+    def get_stages(self):
+        """Method should be overridden in encoder"""
+        raise NotImplementedError
+
+    def make_dilated(self, stage_list, dilation_list):
+        stages = self.get_stages()
+        for stage_indx, dilation_rate in zip(stage_list, dilation_list):
+            utils.replace_strides_with_dilation(
+                module=stages[stage_indx],
+                dilation_rate=dilation_rate,
+            )

segmentation_models_pytorch/segmentation_models_pytorch/encoders/_preprocessing.py

@@ -0,0 +1,23 @@
+import numpy as np
+
+
+def preprocess_input(
+    x, mean=None, std=None, input_space="RGB", input_range=None, **kwargs
+):
+
+    if input_space == "BGR":
+        x = x[..., ::-1].copy()
+
+    if input_range is not None:
+        if x.max() > 1 and input_range[1] == 1:
+            x = x / 255.0
+
+    if mean is not None:
+        mean = np.array(mean)
+        x = x - mean
+
+    if std is not None:
+        std = np.array(std)
+        x = x / std
+
+    return x

segmentation_models_pytorch/segmentation_models_pytorch/encoders/_utils.py

@@ -0,0 +1,50 @@
+import torch
+import torch.nn as nn
+
+
+def patch_first_conv(model, in_channels):
+    """Change first convolution layer input channels.
+    In case:
+        in_channels == 1 or in_channels == 2 -> reuse original weights
+        in_channels > 3 -> make random kaiming normal initialization
+    """
+
+    # get first conv
+    for module in model.modules():
+        if isinstance(module, nn.Conv2d):
+            break
+
+    # change input channels for first conv
+    module.in_channels = in_channels
+    weight = module.weight.detach()
+    reset = False
+
+    if in_channels == 1:
+        weight = weight.sum(1, keepdim=True)
+    elif in_channels == 2:
+        weight = weight[:, :2] * (3.0 / 2.0)
+    else:
+        reset = True
+        weight = torch.Tensor(
+            module.out_channels,
+            module.in_channels // module.groups,
+            *module.kernel_size
+        )
+
+    module.weight = nn.parameter.Parameter(weight)
+    if reset:
+        module.reset_parameters()
+
+
+def replace_strides_with_dilation(module, dilation_rate):
+    """Patch Conv2d modules replacing strides with dilation"""
+    for mod in module.modules():
+        if isinstance(mod, nn.Conv2d):
+            mod.stride = (1, 1)
+            mod.dilation = (dilation_rate, dilation_rate)
+            kh, kw = mod.kernel_size
+            mod.padding = ((kh // 2) * dilation_rate, (kh // 2) * dilation_rate)
+
+            # Kostyl for EfficientNet
+            if hasattr(mod, "static_padding"):
+                mod.static_padding = nn.Identity()

segmentation_models_pytorch/segmentation_models_pytorch/encoders/densenet.py

@@ -0,0 +1,146 @@
+""" Each encoder should have following attributes and methods and be inherited from `_base.EncoderMixin`
+
+Attributes:
+
+    _out_channels (list of int): specify number of channels for each encoder feature tensor
+    _depth (int): specify number of stages in decoder (in other words number of downsampling operations)
+    _in_channels (int): default number of input channels in first Conv2d layer for encoder (usually 3)
+
+Methods:
+
+    forward(self, x: torch.Tensor)
+        produce list of features of different spatial resolutions, each feature is a 4D torch.tensor of
+        shape NCHW (features should be sorted in descending order according to spatial resolution, starting
+        with resolution same as input `x` tensor).
+
+        Input: `x` with shape (1, 3, 64, 64)
+        Output: [f0, f1, f2, f3, f4, f5] - features with corresponding shapes
+                [(1, 3, 64, 64), (1, 64, 32, 32), (1, 128, 16, 16), (1, 256, 8, 8),
+                (1, 512, 4, 4), (1, 1024, 2, 2)] (C - dim may differ)
+
+        also should support number of features according to specified depth, e.g. if depth = 5,
+        number of feature tensors = 6 (one with same resolution as input and 5 downsampled),
+        depth = 3 -> number of feature tensors = 4 (one with same resolution as input and 3 downsampled).
+"""
+
+import re
+import torch.nn as nn
+
+from pretrainedmodels.models.torchvision_models import pretrained_settings
+from torchvision.models.densenet import DenseNet
+
+from ._base import EncoderMixin
+
+
+class TransitionWithSkip(nn.Module):
+
+    def __init__(self, module):
+        super().__init__()
+        self.module = module
+
+    def forward(self, x):
+        for module in self.module:
+            x = module(x)
+            if isinstance(module, nn.ReLU):
+                skip = x
+        return x, skip
+
+
+class DenseNetEncoder(DenseNet, EncoderMixin):
+    def __init__(self, out_channels, depth=5, **kwargs):
+        super().__init__(**kwargs)
+        self._out_channels = out_channels
+        self._depth = depth
+        self._in_channels = 3
+        del self.classifier
+
+    def make_dilated(self, stage_list, dilation_list):
+        raise ValueError("DenseNet encoders do not support dilated mode "
+                         "due to pooling operation for downsampling!")
+
+    def get_stages(self):
+        return [
+            nn.Identity(),
+            nn.Sequential(self.features.conv0, self.features.norm0, self.features.relu0),
+            nn.Sequential(self.features.pool0, self.features.denseblock1,
+                          TransitionWithSkip(self.features.transition1)),
+            nn.Sequential(self.features.denseblock2, TransitionWithSkip(self.features.transition2)),
+            nn.Sequential(self.features.denseblock3, TransitionWithSkip(self.features.transition3)),
+            nn.Sequential(self.features.denseblock4, self.features.norm5)
+        ]
+
+    def forward(self, x):
+
+        stages = self.get_stages()
+
+        features = []
+        for i in range(self._depth + 1):
+            x = stages[i](x)
+            if isinstance(x, (list, tuple)):
+                x, skip = x
+                features.append(skip)
+            else:
+                features.append(x)
+
+        return features
+
+    def load_state_dict(self, state_dict):
+        pattern = re.compile(
+            r"^(.*denselayer\d+\.(?:norm|relu|conv))\.((?:[12])\.(?:weight|bias|running_mean|running_var))$"
+        )
+        for key in list(state_dict.keys()):
+            res = pattern.match(key)
+            if res:
+                new_key = res.group(1) + res.group(2)
+                state_dict[new_key] = state_dict[key]
+                del state_dict[key]
+
+        # remove linear
+        state_dict.pop("classifier.bias")
+        state_dict.pop("classifier.weight")
+
+        super().load_state_dict(state_dict)
+
+
+densenet_encoders = {
+    "densenet121": {
+        "encoder": DenseNetEncoder,
+        "pretrained_settings": pretrained_settings["densenet121"],
+        "params": {
+            "out_channels": (3, 64, 256, 512, 1024, 1024),
+            "num_init_features": 64,
+            "growth_rate": 32,
+            "block_config": (6, 12, 24, 16),
+        },
+    },
+    "densenet169": {
+        "encoder": DenseNetEncoder,
+        "pretrained_settings": pretrained_settings["densenet169"],
+        "params": {
+            "out_channels": (3, 64, 256, 512, 1280, 1664),
+            "num_init_features": 64,
+            "growth_rate": 32,
+            "block_config": (6, 12, 32, 32),
+        },
+    },
+    "densenet201": {
+        "encoder": DenseNetEncoder,
+        "pretrained_settings": pretrained_settings["densenet201"],
+        "params": {
+            "out_channels": (3, 64, 256, 512, 1792, 1920),
+            "num_init_features": 64,
+            "growth_rate": 32,
+            "block_config": (6, 12, 48, 32),
+        },
+    },
+    "densenet161": {
+        "encoder": DenseNetEncoder,
+        "pretrained_settings": pretrained_settings["densenet161"],
+        "params": {
+            "out_channels": (3, 96, 384, 768, 2112, 2208),
+            "num_init_features": 96,
+            "growth_rate": 48,
+            "block_config": (6, 12, 36, 24),
+        },
+    },
+}